Editorial note: This blog was reviewed with ChatGPT 6 Astra to improve readability.
This blog is written for applied economists who want a moderately technical account of a theoretical advance in NARDL estimation. It develops the mathematical intuition through explicit calculations, matrix examples, and explanations of the asymptotic results. It is a reading companion to Jin Seo Cho’s original paper, which remains essential for understanding the full assumptions, proofs, and inference results. The explanations below do not replace that paper.
NARDL has become a widely used tool in applied time-series econometrics. It allows economists to study whether increases and decreases in an explanatory variable have different short-run and long-run effects, while estimating the resulting dynamic equation by ordinary least squares. This combination of flexibility and ease of estimation helps explain its appeal in research on exchange rates, energy prices, trade, and other asymmetric economic relationships.
Cho (2026) fills a gap between this widespread practice and its econometric justification. He establishes the consistency and limiting distribution of the conventional one-step OLS estimator for a NARDL model with a time trend. The difficulty comes from the positive and negative partial sums used to represent asymmetry: they generate deterministic trends that make the usual normalized regressor matrix singular in the limit.
Cho uses Frisch–Waugh–Lovell transformations to separate those trends from the stochastic fluctuations and derive the behavior of the original estimator. This blog explains the estimation argument in Sections 1–4 of the paper. The main text develops the intuition and works through the principal calculations; eight linked appendices explain the supporting algebra and limit calculations in greater detail. The applied discussion also draws on the paper’s comparison of OLS and two-stage NARDL estimation.
Important equations are numbered for reference within this article. Main-text numbers such as (1) and appendix numbers such as (A.1) belong to this blog; references to Cho’s equation numbers are identified explicitly.
1. The econometric foundation for a widely used estimator
The NARDL framework developed by Shin, Yu, and Greenwood-Nimmo (2014) provides a way to estimate asymmetric relationships without specifying an entirely separate dynamic model for every regime. Positive and negative changes are accumulated separately and enter a common regression. Applied researchers can then estimate different responses to increases and decreases in the same explanatory variable.
Once these partial sums have been constructed, OLS estimates the coefficients of a linear regression. To establish consistency and a limiting distribution, however, the proof must track how the regressors and their cross-products behave as the sample grows. In the model studied by Cho, several regressors share a deterministic time trend. The usual normalization makes their limiting columns proportional, so the matrix that a conventional OLS proof would invert is singular.
The stochastic fluctuations around those trends still distinguish the regressors in a finite sample. Cho derives a representation in which those fluctuations remain in the limiting calculation. His contribution establishes the asymptotic behavior of the original one-step OLS coefficients, complementing earlier work on the NARDL framework and two-stage estimation.
The method builds on the projection arguments of Park and Phillips (1988, 1989) for regressions with integrated processes. Exact FWL identities connect the original regression to regressions with nonsingular limiting matrices. The proof then reconstructs every original coefficient and shows why different estimation errors determine their convergence rates.
For applied work, this gives a theoretical justification for a familiar estimator and a basis for interpreting its precision. A coefficient in the dynamic equation, a long-run slope, and an adjustment speed are related parameters, each with its own role in the asymptotic argument. The distinction becomes visible once the regression is written in error-correction form.
Main takeaway. Cho establishes the large-sample properties of conventional one-step OLS for a NARDL model with a time trend. The proof addresses the deterministic drifts introduced by the positive and negative partial sums.
Calculations: Appendix C identifies the singularity; Appendix D derives the projection argument.
2. From economic asymmetry to a linear regression
For one explanatory variable, split each observed change into its positive and negative parts:
\[\Delta x_t^+=\max(\Delta x_t,0),\qquad \Delta x_t^-=\min(\Delta x_t,0).\]A change of \(-3\) enters the negative component as \(-3\). Accumulate the two series separately:
Equation (1) records the history of each sign separately. The original cumulative change is their sum, \(x_t-x_0=x_t^++x_t^-\). Keeping the two components separate allows increases and decreases to have different coefficients. If an energy price rises by one unit and subsequently falls by one unit, its original level returns to its starting point. The two partial sums still record a cumulative increase of one unit and a cumulative decrease of one unit. A NARDL model can therefore distinguish the effects of the two movements.
Now suppose there are \(k\) original explanatory variables. Each variable produces two accumulated series: one for its positive changes and one for its negative changes. There are therefore \(k\) positive partial sums and \(k\) negative partial sums, making \(k+k=2k\) entries altogether. We put these entries one below another in a single column. This is why the combined vector has \(2k\) rows and one column, written \(2k\times1\). The order of a dimension is always rows first, columns second.
Let \(x_{i,t}^+\) and \(x_{i,t}^-\) denote the two partial sums for original variable \(i\), where \(i=1,\ldots,k\). The positive and negative columns, and the column formed by stacking them, are
In Equation (2), each of the first two columns has \(k\) rows. In the last column, the first \(k\) rows contain positive partial sums and the next \(k\) rows contain negative partial sums. Stacking adds rows; it does not add a second column. For example, with two original variables, \(k=2\), the combined column is
\[ \ddot x_t= \begin{pmatrix}x_{1,t}^+\\x_{2,t}^+\\x_{1,t}^-\\x_{2,t}^-\end{pmatrix}, \qquad 2k\times1=(2\times2)\times1=4\times1. \]The compact notation \(\ddot x_t=(x_t^{+\prime},x_t^{-\prime})’\) means exactly the same stacking operation. A prime denotes a transpose: the inner primes turn the two columns into rows; placing those rows side by side makes a single row with \(2k\) entries; the outer prime turns that row back into a column. The double dot is a notation label, not a derivative. Lowercase \(k\) counts the original explanatory variables. Uppercase \(K\), introduced below, counts all regressors, including the lagged outcome, trend, constant, and short-run terms.
The model is nonlinear in the original changes because their sign determines which component changes. Once those components have been constructed, the regression is linear in its unknown coefficients.
The error-correction representation in Equation (3) collects the resulting regressors:
The vector \(z_{2t}\) contains lagged changes in \(y_t\) and the current and lagged positive and negative changes in the explanatory variables. Their coefficients describe short-run dynamics. The lagged levels and the time trend describe the longer-run relationship. A star marks a population parameter. Here \(q_t\) is the time regressor; the unsubscripted \(q\) in NARDL\((p,q)\) is the explanatory-variable lag order.
Counting variables and reading dimensions. A dimension gives the number of rows followed by the number of columns. A \(2k\times1\) vector has \(2k\) rows and one column; a scalar is a single number. The following counts will be used throughout the calculations.
| Symbol | What it counts |
|---|---|
| \(T\) | Observations used in the regression. |
| \(k\) | Original explanatory variables, giving \(2k\) positive and negative partial sums. |
| \(p\), \(q\) | Dependent-variable and explanatory-variable lag orders, respectively. |
| \(s=(p-1)+2kq\) | Short-run regressors: \(p-1\) lagged outcome changes and \(2kq\) current and lagged signed explanatory-variable changes. |
| \(d=2k+2\) | Regressors in the drifting block: one lagged outcome, \(2k\) partial sums, and one time trend. |
| \(K=d+s+1=2k+s+3\) | All regressors: the drifting block, \(s\) short-run regressors, and one constant. |
Table notes. Read each row as a count of observations, variables, or regressors. For \(k=1\), \(p=2\), and \(q=1\), there are \(s=1+2=3\) short-run regressors, \(d=4\) drifting regressors, and \(K=8\) regressors altogether. Increasing the sample size \(T\) adds observations while these counts remain fixed for this specification. For each original explanatory variable, the short-run block contains \(q\) positive changes and \(q\) negative changes, dated \(t,t-1,\ldots,t-q+1\). Across \(k\) variables, this gives \(2kq\) entries. Adding the \(p-1\) lagged changes in the dependent variable gives \(s=(p-1)+2kq\).
Performing the multiplication. With one original explanatory variable, \(k=1\), there are two partial sums and two corresponding coefficients. Write the entries before multiplying:
\[ \theta_*= \begin{pmatrix}\theta_*^+\\\theta_*^-\end{pmatrix}, \qquad \ddot x_{t-1}= \begin{pmatrix}x_{t-1}^+\\x_{t-1}^-\end{pmatrix}. \]Transposing the coefficient column makes a row. Multiply its first entry by the first regressor, multiply its second entry by the second regressor, and add:
\[ \begin{aligned} \theta_*^{\prime}\ddot x_{t-1} &=\begin{pmatrix}\theta_*^+&\theta_*^-\end{pmatrix} \begin{pmatrix}x_{t-1}^+\\x_{t-1}^-\end{pmatrix}\\ &=\theta_*^+x_{t-1}^++\theta_*^-x_{t-1}^-. \end{aligned} \]The result is one number, called a scalar. For illustrative coefficient values \(0.8,0.4\) and partial sums \(5,-3\), the same operation gives
\[ \begin{pmatrix}0.8&0.4\end{pmatrix} \begin{pmatrix}5\\-3\end{pmatrix} =0.8(5)+0.4(-3)=4-1.2=2.8. \]These values illustrate the multiplication. The negative partial sum keeps its negative sign. For \(k\) original variables, the operation adds \(2k\) such products: a \(1\times2k\) row times a \(2k\times1\) column gives one scalar.
The short-run product works in the same way. In the table’s example \(p=2,q=1,k=1\), the three short-run regressors and their coefficients are
\[ z_{2t}=\begin{pmatrix}\Delta y_{t-1}\\\Delta x_t^+\\\Delta x_t^-\end{pmatrix}, \qquad \kappa_*=\begin{pmatrix}\phi_{1*}\\\pi_{0*}^+\\\pi_{0*}^-\end{pmatrix}. \] \[ \begin{aligned} \kappa_*^{\prime}z_{2t} &=\begin{pmatrix}\phi_{1*}&\pi_{0*}^+&\pi_{0*}^-\end{pmatrix} \begin{pmatrix}\Delta y_{t-1}\\\Delta x_t^+\\\Delta x_t^-\end{pmatrix}\\ &=\phi_{1*}\Delta y_{t-1} +\pi_{0*}^+\Delta x_t^+ +\pi_{0*}^-\Delta x_t^-. \end{aligned} \]The symbols \(\phi_{1*},\pi_{0*}^+,\pi_{0*}^-\) name the coefficients on these three changes. Their relation to the coefficients in the original levels equation is derived in Appendix A. The four drifting regressors are \(y_{t-1},x_{t-1}^+,x_{t-1}^-,q_t\). Adding the constant and these three short-run regressors gives all eight columns counted in the table. Every contribution to the equation for the scalar \(\Delta y_t\) is now an explicit coefficient times a regressor.
The coefficients on lagged partial sums in this dynamic equation are collected in the \(2k\times1\) vector \(\theta_*\). The equilibrium slopes form another \(2k\times1\) vector, \(\beta_*\), obtained by dividing every component by the scalar adjustment coefficient. The equilibrium trend coefficient \(\zeta_*\) is scalar:
Applying Equation (4), \(\rho_*=-0.4\) and \(\theta_*^+=0.8\) imply an equilibrium slope \(\beta_*^+=2\). The coefficient \(0.8\) appears in the equation for the next change in the outcome; the coefficient \(2\) describes the equilibrium association between the outcome and the positive partial sum.
The long-run relationship is
The product \(\beta_*’\ddot x_{t-1}\) is scalar, and \(\zeta_*\), \(\nu_*\), and \(u_{t-1}\) are scalars as well. The cointegration error \(u_{t-1}\) in Equation (5) measures the outcome’s deviation from this relationship. The intercept \(\nu_*\) centers that error at zero. Substituting Equation (5) into Equation (3) gives the adjustment equation:
In Equation (6), if the outcome lies two units above its equilibrium relation and \(\rho_*=-0.4\), the adjustment term contributes \(-0.8\) to its next change. The other terms in the equation determine the remainder of that change. This is the economic role of the adjustment coefficient whose estimation error becomes important later.
Asymmetry concerns the equilibrium slopes as well as the short-run coefficients. In a numerical illustration with \(\beta_*^+=2\) and \(\beta_*^-=1\), a one-unit increase in the explanatory variable raises the equilibrium component by two units. A one-unit decrease lowers it by one unit because the negative partial sum changes by \(-1\). These examples describe the model’s arithmetic.
The assumptions used in the paper
Cho’s Assumption 1 maintains cointegration and stable dynamics, a nonzero mean change \(\delta_*=E(\Delta y_t)\), stationary short-run components with suitable moment and dependence conditions, and the required nonsingular covariance matrices. The dynamic innovation has conditional mean zero and constant conditional variance relative to an information set that includes contemporaneous explanatory variables. The time trend is included as a regressor, including when its true coefficient is zero.
Main takeaway. The error-correction model separates adjustment toward equilibrium from short-run dynamics. Its level coefficients \(\theta_*\) and long-run slopes \(\beta_*=-\theta_*/\rho_*\) answer different economic questions, so their estimates and convergence rates must be distinguished.
Calculations and symbol definitions: Appendix A.
3. Why partial sums generate trends
Separating a change into positive and negative components changes the means of the resulting series. Consider a simple illustration in which \(\Delta x_t\) takes the values \(-2,-1,+1,+2\), each with probability one quarter. The original mean is zero. Its positive part takes the values \(0,0,1,2\), while its negative part takes the values \(-2,-1,0,0\). Therefore,
\[ E(\Delta x_t^+)=\frac{0+0+1+2}{4}=\frac34, \qquad E(\Delta x_t^-)=-\frac34. \]Why the expected partial sum contains a factor \(t\). The positive partial sum adds the positive changes observed up to date \(t\):
\[ x_t^+=\Delta x_1^++\Delta x_2^++\cdots+\Delta x_t^+. \]There are \(t\) terms. Expectation distributes over addition, and every term in this example has the same expectation \(3/4\). Therefore,
\[ \begin{aligned} E(x_t^+) &=E(\Delta x_1^+)+E(\Delta x_2^+)+\cdots+E(\Delta x_t^+)\\ &=\underbrace{\frac34+\frac34+\cdots+\frac34}_{t\text{ terms}}\\ &=t\times\frac34=\frac{3t}{4}. \end{aligned} \]The negative partial sum adds \(t\) terms whose common expectation is \(-3/4\):
\[ E(x_t^-) =\underbrace{\left(-\frac34\right)+\cdots+\left(-\frac34\right)}_{t\text{ terms}} =-\frac{3t}{4}. \]After four periods, the expected positive sum is \(3\) and the expected negative sum is \(-3\). These are expectations over possible realizations of the changes; an observed four-period sum can differ. Linearity of expectation gives the calculation without requiring independence across dates.
Here lowercase \(t\) counts the increments accumulated by a particular date. Uppercase \(T\) denotes the length of the full sample; at the last date, \(t=T\). The deterministic parts cancel in the combined series because \(3t/4-3t/4=0\). The NARDL regression includes the two partial sums separately, so their individual linear drifts remain in its regressors.
To distinguish drift from fluctuations, define the mean vector \(\mu_*:=E(\Delta\ddot x_t)\), the centered increments \(s_j=\Delta\ddot x_j-\mu_*\), and their cumulative sum \(m_{t-1}=\sum_{j=1}^{t-1}s_j\). Each is a \(2k\times1\) column vector, with one entry for each signed component. The increment vector \(s_j\) is distinct from the scalar count \(s\) in the table above. Adding the mean back to each increment gives the exact decomposition
This decomposition follows by adding back the same mean at every date:
\[ \begin{aligned} \ddot x_{t-1} &=\sum_{j=1}^{t-1}\Delta\ddot x_j =\sum_{j=1}^{t-1}(\mu_*+s_j)\\ &=\underbrace{\mu_*+\cdots+\mu_*}_{t-1\text{ copies}} +\sum_{j=1}^{t-1}s_j\\ &=(t-1)\mu_*+m_{t-1}=q_t\mu_*+m_{t-1}. \end{aligned} \]The first term is the accumulated mean change. The second adds the departures of individual changes from that mean. Its expectation is zero, but its realized value can be substantial.
The size of those departures can be understood from one scalar component. For this illustration, let \(\eta_j\) denote one component of the vector \(s_j\), and let \(\mu\) denote the corresponding scalar mean increment. Suppose the centered increments \(\eta_j\) are independent, each with variance \(\sigma^2\). The variance of their sum equals the sum of their variances because the cross-covariances are zero:
\[ \operatorname{Var}\!\left(\sum_{j=1}^{t}\eta_j\right) =\sum_{j=1}^{t}\operatorname{Var}(\eta_j) =t\sigma^2. \]Standard deviation is the square root of variance. Consequently,
\[ \operatorname{SD}\!\left(\sum_{j=1}^{t}\eta_j\right) =\sigma\sqrt t. \]For unit-variance increments, 100 observations give variance 100 and standard deviation 10. With 400 observations, the variance rises to 400 and the standard deviation to 20. Quadrupling the number of increments doubles the typical fluctuation. A deterministic drift, in contrast, quadruples over the same interval.
The relative size of the stochastic fluctuation therefore decreases:
\[ \frac{\sigma\sqrt t}{|\mu|t} =\frac{\sigma}{|\mu|\sqrt t}\longrightarrow0 \qquad(\mu\ne0). \]The fluctuations grow in absolute size while shrinking relative to the trend. The paper obtains the corresponding square-root scale under its assumptions on dependence; independence is used here to make the variance calculation transparent.
Reading the powers of the sample size
In an asymptotic argument, we imagine longer samples generated by the same model, keeping its population parameters and lag orders fixed. We ask how the quantities in the calculation change as the number of observations grows.
Begin with three ordinary quantities: \(6\), \(6T\), and \(6T^2\). Increasing the sample length from \(T=100\) to \(T=200\) leaves the first at \(6\), doubles the second from \(600\) to \(1{,}200\), and quadruples the third from \(60{,}000\) to \(240{,}000\). The powers tell us how the quantities respond when the sample grows. The multiplier \(6\) changes their level without changing that response.
Big O records an upper bound on this growth. We write \(6=O(1)\), \(6T=O(T)\), and \(6T^2=O(T^2)\). The first statement does not say that six equals one: it says that six stays on a constant scale as \(T\) increases. Divide each quantity by its proposed scale:
\[ \frac{6}{1}=6,\qquad \frac{6T}{T}=6,\qquad \frac{6T^2}{T^2}=6. \]Every ratio stays bounded. More generally, for a positive scale \(a_T\), the statement \(f_T=O(a_T)\) means that some fixed constant \(C\) bounds the absolute ratio for all sufficiently large \(T\):
The constant \(C\) in Equation (8) can be large, but it cannot increase with \(T\). For example, \((3T^2+2T)/T^2=3+2/T\leq5\) for \(T\geq1\), so \(3T^2+2T=O(T^2)\). The linear term becomes small relative to the quadratic term. This is the ratio argument developed in my earlier post, Why is Big O of a constant always equal to O(1)? Here the limit is always an increasing sample size, \(T\to\infty\).
An upper bound need not be the tightest one: \(T=O(T^2)\) is also true, since \(T/T^2=1/T\) stays bounded. Little o records the stronger statement that the ratio disappears. Thus \(T=o(T^2)\), because \(1/T\to0\). The quantity \(T\) itself grows; it becomes negligible only relative to \(T^2\). Likewise, \(6/T=o(1)\), whereas the constant \(6\) is \(O(1)\) but is not \(o(1)\).
These examples also explain the notation used for individual dates. The mean drift \(3t/4\) is \(O(t)\); its square \(9t^2/16\) is \(O(t^2)\). At a date that stays, say, halfway through an expanding sample, \(t\) is approximately \(T/2\). The corresponding scales are then \(T\) and \(T^2\). This reasoning concerns dates moving outward with the sample; a fixed early date does not grow when more observations are added at the end.
What convergence in probability means
Estimates and sample averages vary across possible samples. To describe their accuracy, imagine repeatedly drawing a sample of length \(T\) from the same model, calculating a quantity \(X_T\) each time, and comparing it with a fixed target \(c\). Choose a tolerance, for example \(0.01\). Convergence in probability means that the chance of missing the target by more than this tolerance tends to zero as the sample length increases. The requirement must hold for every fixed positive tolerance, however small:
In Equation (9), \(|X_T-c|\) is the size of the error, \(\varepsilon\) is the acceptable error, and \(\Pr\) measures the chance of exceeding it across samples generated by the model. For an estimator, convergence to the true parameter is called consistency. It does not require each additional observation to move an estimate closer to the truth.
For a simple example, suppose independent observations have mean \(c\) and variance \(\sigma^2\). Their sample average \(\overline X_T\) has variance \(\sigma^2/T\). Chebyshev’s inequality translates this shrinking variance into a bound on the chance of a large error:
\[ \Pr\!\left(|\overline X_T-c|>\varepsilon\right) \leq\frac{\sigma^2}{T\varepsilon^2}\longrightarrow0. \]For any chosen tolerance, the upper bound falls as \(T\) grows. Thus \(\overline X_T\xrightarrow{p}c\). This elementary example explains the meaning of the arrow; the NARDL results require the paper’s own assumptions about dynamics and dependence.
Reading \(O_p\) and \(o_p\)
Random quantities need a probabilistic version of the growth bound. Return to the centered cumulative sum discussed above. With independent unit-variance increments, its standard deviation is \(\sqrt T\). Dividing the sum by \(\sqrt T\) produces a quantity with standard deviation one. Individual realizations can still be unusually large, but the rescaled variation does not spread without bound as the sample grows. This motivates the notation \(O_p(\sqrt T)\).
More precisely, \(X_T=O_p(a_T)\) means that the ratio \(X_T/a_T\) is bounded in probability. For any desired probability level, such as 99%, a sufficiently wide fixed interval contains that ratio with at least that probability for all sufficiently large samples. The bound can depend on how high a probability we require; it does not grow with \(T\). This does not bound every possible realization.
The notation \(X_T=o_p(a_T)\) says more: \(X_T/a_T\xrightarrow{p}0\). The ratio eventually lies within any fixed positive tolerance of zero with probability approaching one. In particular, \(o_p(1)\) is a random remainder that vanishes in probability.
For a concrete comparison, let \(U\) have a standard normal distribution and define two hypothetical estimation errors:
\[ X_T=\frac{U}{\sqrt T},\qquad Y_T=\frac{U}{T}. \]Their standard deviations are \(1/\sqrt T\) and \(1/T\). At \(T=100\), those are \(0.1\) and \(0.01\); at \(T=400\), they are \(0.05\) and \(0.0025\). Both errors shrink, but the second shrinks faster. Compare them using the same multiplier:
\[ \sqrt T X_T=U,\qquad \sqrt T Y_T=\frac{U}{\sqrt T}\xrightarrow{p}0. \]The first scaled error retains its normal variation; the second disappears. Hence \(X_T=O_p(T^{-1/2})\), while \(Y_T=o_p(T^{-1/2})\) and \(Y_T=O_p(T^{-1})\). As with deterministic Big O, an \(O_p\) statement gives an upper bound: a quantity that shrinks faster also satisfies the bound. A limiting distribution with nonzero variation shows that the chosen scale preserves random fluctuations; an \(O_p\) bound alone does not. We will meet those distributions in Section 7.
Why the squared trend sum has a \(T^3\) scale. It helps to distinguish a fixed date from a date that moves with the sample. At date \(t=5\), the trend is always \(q_5=4\), even if we extend the sample. At the midpoint of a sample of \(100\) observations, however, \(q_{50}=49\) and \(q_{50}^2=2{,}401\). At the midpoint of a sample of \(200\), \(q_{100}=99\) and \(q_{100}^2=9{,}801\). Doubling the sample length approximately doubles the trend at the midpoint and approximately quadruples its square. This is the meaning of the scales \(T\) and \(T^2\) for observations at a fixed fraction of the sample.
Now add the squared trend values over all dates. First factor \(T^2\) out of each squared observation; then write the remaining sum as its average multiplied by the number of observations:
\[ \begin{aligned} \sum_{t=1}^{T}q_t^2 &=T^2\sum_{t=1}^{T}\left(\frac{q_t}{T}\right)^2\\ &=T^2\times T\times \left[\frac1T\sum_{t=1}^{T}\left(\frac{q_t}{T}\right)^2\right]. \end{aligned} \]The first factor, \(T^2\), comes from the squared trend values. The additional factor, \(T\), comes from the identity sum = number of observations × average. It is introduced together with \(1/T\) inside the brackets, so the equality is exact. The bracket is the average of the squared relative dates \(0/T,1/T,\ldots,(T-1)/T\); it tends to \(1/3\). The exact sum confirms the result:
\[ \sum_{t=1}^{T}q_t^2=\frac{(T-1)T(2T-1)}6, \qquad \frac1{T^3}\sum_{t=1}^{T}q_t^2 =\frac13-\frac1{2T}+\frac1{6T^2}\longrightarrow\frac13. \]| Sample length \(T\) | \(\sum_{t=1}^Tq_t^2\) | \(T^{-3}\sum_{t=1}^Tq_t^2\) |
|---|---|---|
| 10 | 285 | 0.285 |
| 100 | 328,350 | 0.32835 |
| 1,000 | 332,833,500 | 0.3328335 |
Table notes. Each row uses the trend column \(0,1,\ldots,T-1\). The second column adds the squares of all its entries; the third divides that sum by \(T^3\). The raw sum increases rapidly, while its normalized value approaches \(1/3\). These numbers are exact calculations from the time indices, so this convergence is deterministic. The corresponding column divisor is \(T^{3/2}\), because squaring \(q_t/T^{3/2}\) puts \(T^3\) in the denominator. Appendix B.3 derives the sum formula without skipping the intermediate steps.
For one scalar component, write \(M_{t-1}:=\sum_{j=1}^{t-1}\eta_j\). The variance calculation for independent increments gives \(E(M_{t-1}^2)=\sigma^2(t-1)\). Hence
\[ E\!\left(\sum_{t=1}^{T}M_{t-1}^2\right) =\sigma^2\bigl[0+1+\cdots+(T-1)\bigr] =\frac{\sigma^2T(T-1)}2. \]This expected sum has order \(T^2\). The paper’s functional limit assumptions establish the corresponding stochastic scale. The trend contribution therefore grows faster: dividing both sums by \(T^3\) preserves the trend contribution while shrinking the fluctuation contribution by a factor of order \(1/T\). For signed sums such as regressor–error products, cancellation matters; their rates require a separate calculation, given in Appendix B.4.
Main takeaway. Separately accumulating positive and negative changes creates deterministic trends even when the original change has mean zero. These trends grow linearly with time, while the centered cumulative fluctuations have a square-root scale.
The variance calculation, numerical examples, and exact sums: Appendix B.
4. Where the usual OLS argument breaks down
The difficulty becomes concrete in the preceding example. The two partial sums have the representations \(x_{t-1}^+=(3/4)q_t+m_{t-1}^+\) and \(x_{t-1}^-=-(3/4)q_t+m_{t-1}^-\). Retaining only their dominant deterministic parts would give
\[ \theta_*^+x_{t-1}^+ +\theta_*^-x_{t-1}^-+\xi_*q_t \approx \left(\frac34\theta_*^+-\frac34\theta_*^-+\xi_*\right)q_t. \]At this leading scale, the three regressors explain the same time pattern. Only the coefficient combination in parentheses is distinguishable. Different values of the three coefficients can produce the same combination. The fluctuations \(m_{t-1}^+\) and \(m_{t-1}^-\) provide additional variation that can distinguish their separate effects in the observed sample.
The outcome also contains a deterministic drift:
\[ y_{t-1} =\delta_*q_t+\sum_{j=1}^{t-1}(\Delta y_j-\delta_*), \qquad \delta_*=E(\Delta y_t), \]where initial values are suppressed. Collect the trending regressors and their drift coefficients:
\[ v_t= \begin{pmatrix} y_{t-1}\\ \ddot x_{t-1}\\ q_t \end{pmatrix}, \qquad a= \begin{pmatrix} \delta_*\\ \mu_*\\ 1 \end{pmatrix}. \]Both \(v_t\) and \(a\) are \(d\times1\) column vectors, where \(d=1+2k+1=2k+2\): their displayed middle block contains \(2k\) entries. The vector \(a\) contains one drift coefficient for each regressor in \(v_t\). Its last entry is one because the explicit time trend is exactly \(1\times q_t\). Thus, \(aq_t\) is also \(d\times1\) and contains the deterministic trend in every regressor. Define the \(d\times1\) vector \(r_t\) as the departures from those trends:
\[ v_t=aq_t+r_t, \qquad r_t= \begin{pmatrix} y_{t-1}-\delta_*q_t\\ \ddot x_{t-1}-\mu_*q_t\\ 0 \end{pmatrix}. \]The last departure is zero because the explicit trend has no random component. The other entries contain the fluctuations discussed in Section 3. Under the paper’s functional limit assumptions, their stochastic magnitude is at most \(\sqrt T\) at times proportional to sample size, whereas a nonzero deterministic drift grows like \(T\).
Multiplying the trend components. The transpose \(a’\) is a \(1\times d\) row vector. Multiplying the \(d\times1\) column \(a\) by that row produces the \(d\times d\) matrix \(aa’\): each row is paired with every column. Since \(q_t\) is a scalar,
\[ (aq_t)(aq_t)’=(aq_t)(q_ta’)=q_t^2aa’. \]The matrices \(v_tv_t’\) and \(aa’\), their normalized sums, and the limiting matrix \(aa’/3\) all have dimension \(d\times d\). The matrix \(aa’\) contains all pairwise products of the drift coefficients. Its entry in row \(i\), column \(j\) is \(a_i a_j\), with \(i,j=1,\ldots,d\). For example, multiplying the positive partial sum’s trend \(\mu_*^+q_t\) by the explicit trend \(1\times q_t\) gives \(\mu_*^+q_t^2\).
For \(k=1\), the middle block contains the positive and negative partial sums. Writing every entry makes this multiplication visible:
\[ a=\begin{pmatrix}\delta_*\\\mu_*^+\\\mu_*^-\\1\end{pmatrix}, \qquad a’=\begin{pmatrix}\delta_*&\mu_*^+&\mu_*^-&1\end{pmatrix}, \] \[ aa’= \begin{pmatrix} \delta_*^2&\delta_*\mu_*^+&\delta_*\mu_*^-&\delta_*\\ \mu_*^+\delta_*&(\mu_*^+)^2&\mu_*^+\mu_*^-&\mu_*^+\\ \mu_*^-\delta_*&\mu_*^-\mu_*^+&(\mu_*^-)^2&\mu_*^-\\ \delta_*&\mu_*^+&\mu_*^-&1 \end{pmatrix}. \]The product is \(4\times4\): each of the four entries in the column multiplies each of the four entries in the row. For example, row 2, column 3 pairs the positive and negative partial sums. Multiplying their trend components gives \((\mu_*^+q_t)(\mu_*^-q_t)=\mu_*^+\mu_*^-q_t^2\). The same factor \(q_t^2\) appears in every position, which explains the matrix identity \((aq_t)(aq_t)’=q_t^2aa’\).
Accounting for the fluctuations. The observed cross-product also contains the departures \(r_t\). For any two components, write \(v_{it}=a_iq_t+r_{it}\) and \(v_{jt}=a_jq_t+r_{jt}\). Expanding their product gives four terms:
\[ v_{it}v_{jt} =a_i a_jq_t^2 +a_iq_tr_{jt} +a_jq_tr_{it} +r_{it}r_{jt}. \]The first is the product of two deterministic trends. The middle two combine a trend with a fluctuation. The last multiplies two fluctuations. Under the paper’s functional limit assumptions, \(\sum_t r_{it}^2\) has at most order \(T^2\), as explained in Appendix B, while \(\sum_t q_t^2\) has order \(T^3\).
The mixed terms can be bounded using the Cauchy–Schwarz inequality, which bounds a sum of products by the product of the two square-root sums of squares:
\[ \left|\sum_{t=1}^{T}q_tr_{it}\right| \leq \left(\sum_{t=1}^{T}q_t^2\right)^{1/2} \left(\sum_{t=1}^{T}r_{it}^2\right)^{1/2}. \]The first factor has order \(T^{3/2}\), and the second has at most order \(T\). Their product therefore has at most order \(T^{5/2}\). After division by \(T^3\), this bound has at most order \(T^{-1/2}=1/\sqrt T\), which tends to zero. The same inequality bounds \(\left|\sum_t r_{it}r_{jt}\right|\) by a product of two terms of at most order \(T\). That product has at most order \(T^2\), so division by \(T^3\) leaves at most order \(1/T\), which also tends to zero.
Consequently, only the product of the deterministic trends survives this normalization. In matrix form,
\[ \frac1{T^3}\sum_{t=1}^{T}v_tv_t’ =aa’\left[\frac1{T^3}\sum_{t=1}^{T}q_t^2\right]+\mathcal{E}_T, \qquad \mathcal{E}_T\xrightarrow{p}0. \]Here the \(d\times d\) matrix \(\mathcal{E}_T\) collects the normalized mixed and fluctuation products just examined. Its entries are random because they depend on the observed sample. Apply the definition of convergence in probability with target zero. For any entry \(\mathcal{E}_{ij,T}\) and every fixed positive tolerance \(\varepsilon\), the statement is
\[ \Pr\!\left(|\mathcal{E}_{ij,T}|>\varepsilon\right)\longrightarrow0. \]The probability refers to samples generated by the model. As sample size grows, an appreciable remaining contribution becomes less likely. For example, a tolerance of \(0.01\) means that the probability of a remainder exceeding \(0.01\) in absolute value tends to zero. The same claim holds for every smaller fixed positive tolerance.
This disappearance concerns the products after division by \(T^3\). The cumulative fluctuations themselves can keep growing, and their unnormalized cross-products can keep growing too. Their growth is slower than the denominator, so they become negligible on the scale used for this matrix limit.
Obtaining the factor \(1/3\). Because \(q_t=t-1\), its squared values are \(0^2,1^2,\ldots,(T-1)^2\). Their exact sum is derived step by step in Appendix B.3. Substituting that formula and dividing each factor by the appropriate power of \(T\) gives
\[ \begin{aligned} \frac1{T^3}\sum_{t=1}^{T}q_t^2 &=\frac{(T-1)T(2T-1)}{6T^3}\\ &=\frac16\left(\frac{T-1}{T}\right)\left(\frac{2T-1}{T}\right)\\ &=\frac16\left(1-\frac1T\right)\left(2-\frac1T\right)\\ &\longrightarrow\frac16\times1\times2=\frac13. \end{aligned} \]The vector \(a\) contains fixed population drift coefficients, so \(aa’\) can be taken outside the sum. Combining the limit \(1/3\) with the vanishing remainder gives
Equation (10) explains each entry of the limiting matrix: multiply the drift coefficient of its row by the drift coefficient of its column, then divide by three. The trend’s diagonal entry is \(1\times1/3=1/3\). In the scalar illustration, the entry joining the positive partial sum and the trend is \((3/4)\times1/3=1/4\).
To read the probability statement concretely, call the normalized positive-partial-sum–trend entry \(C_T\). In the example with \(\mu_*^+=3/4\),
\[ C_T:=\frac1{T^3}\sum_{t=1}^{T}x_{t-1}^+q_t \xrightarrow{p}\frac14. \]Different samples generally produce different values of \(C_T\). Convergence to \(1/4=0.25\) means that these values become concentrated around \(0.25\). Taking a tolerance of \(0.02\) gives
\[ \Pr\!\left(|C_T-0.25|>0.02\right)\longrightarrow0. \]Thus, the chance of a value outside \([0.23,0.27]\) tends to zero. The numbers \(0.02\), \(0.23\), and \(0.27\) illustrate the meaning of the limit; they are not estimated confidence bounds. The result also applies to any narrower fixed tolerance around \(0.25\). The deterministic trend–trend entry needs only an ordinary arrow because its value depends on \(T\) alone; entries involving observed partial sums require a probability statement.
Why this matrix has rank one. Rank counts the number of linearly independent columns. Column \(j\) of \(aa’/3\) is the vector \(a\) multiplied by the single number \(a_j/3\):
\[ \text{column }j\text{ of }\frac13aa’=\frac{a_j}{3}\,a. \]In particular, the last column is \(a/3\), because the last entry of \(a\) is one. Every other column is its own drift coefficient times that last column. The last column is nonzero, and all the columns lie along this one direction. There is therefore exactly one independent column.
The numerical example from Section 3 makes the dependence visible. With \(k=1\), the full drifting block has four entries. Restrict attention here to its last three entries: the two partial sums and the explicit trend. Their drift coefficients are \(3/4\), \(-3/4\), and \(1\). Write this \(3\times1\) subvector as \(a_{\mathrm{sub}}\). First multiply the column by its row transpose:
\[ \begin{aligned} a_{\mathrm{sub}}a_{\mathrm{sub}}’ &=\begin{pmatrix}3/4\\-3/4\\1\end{pmatrix} \begin{pmatrix}3/4&-3/4&1\end{pmatrix}\\ &=\begin{pmatrix} 9/16&-9/16&3/4\\ -9/16&9/16&-3/4\\ 3/4&-3/4&1 \end{pmatrix}. \end{aligned} \]For example, the upper-left entry is \((3/4)(3/4)=9/16\), while the upper-middle entry is \((3/4)(-3/4)=-9/16\). Now divide every entry by three, using the limit of the normalized squared trend sum:
\[ \frac13a_{\mathrm{sub}}a_{\mathrm{sub}}’ =\begin{pmatrix} 3/16&-3/16&1/4\\ -3/16&3/16&-1/4\\ 1/4&-1/4&1/3 \end{pmatrix}. \]Thus \((9/16)/3=9/48=3/16\), and \((3/4)/3=1/4\). To check the column dependence directly, multiply the third column by the two partial-sum drift coefficients:
\[ \frac34\begin{pmatrix}1/4\\-1/4\\1/3\end{pmatrix} =\begin{pmatrix}3/16\\-3/16\\1/4\end{pmatrix}, \qquad -\frac34\begin{pmatrix}1/4\\-1/4\\1/3\end{pmatrix} =\begin{pmatrix}-3/16\\3/16\\-1/4\end{pmatrix}. \]The first result is exactly the first column; the second is exactly the second column. All three therefore point along one independent direction. Adding the lagged outcome gives the same relation: its limiting column is \(\delta_*\) times the trend column. These proportional columns record the common deterministic time pattern in the regressors.
This is the source of the singularity. At the \(T^3\) scale, the normalized matrix retains the common deterministic trend. The smaller random fluctuations that help distinguish the regressors in the observed sample disappear from this limit. Adding the constant and stationary regressors preserves the proportionality among the drifting-regressor columns throughout the full normalized matrix; Appendix C displays those additional blocks.
A singular limit can coexist with a finite-sample inverse. For example,
\[ B_T= \begin{pmatrix}1&1\\1&1+1/T\end{pmatrix}, \qquad \det(B_T)=\frac1T>0. \]The determinant multiplies the two diagonal entries and subtracts the product of the two off-diagonal entries:
\[ \det(B_T)=1\left(1+\frac1T\right)-1(1) =1+\frac1T-1=\frac1T. \]Here \(B_T\) is a \(2\times2\) illustrative matrix, and its determinant is a scalar. It is invertible for every finite positive \(T\), although its limit has identical columns. The NARDL problem likewise concerns the limiting normalized matrix; it does not require exact collinearity in the observed regression.
The small term \(1/T\) is essential to this example: subtracting the first column from the second gives \((0,1/T)’\). It is the entire difference between the columns. Taking the limit removes that difference. To see how it affects a coefficient calculation, solve the illustrative system \(B_Tc_T=b_T\), where \(c_T=(c_{1T},c_{2T})’\) and \(b_T=(b_{1T},b_{2T})’\):
\[ \begin{aligned} c_{1T}+c_{2T}&=b_{1T},\\ c_{1T}+\left(1+\frac1T\right)c_{2T}&=b_{2T}. \end{aligned} \]Subtract the first equation from the second. The \(c_{1T}\) terms cancel, and the difference between the two coefficients on \(c_{2T}\) is \(1/T\):
\[ \frac{c_{2T}}T=b_{2T}-b_{1T}, \qquad c_{2T}=T(b_{2T}-b_{1T}), \qquad c_{1T}=b_{1T}-c_{2T}. \]The solution depends jointly on the small difference between the matrix columns and the small difference between the right-hand entries. For example, if \(b_{2T}-b_{1T}=c_0/T\) for a fixed scalar \(c_0\), then \(c_{2T}=c_0\) at every sample size. The growing entries of an inverse alone therefore do not establish that a coefficient diverges. This is an algebraic illustration, not the NARDL matrix itself. It explains why the proof must keep the smaller terms in both the regressor matrix and the regressor–error products before taking limits.
Let \(z_t\) denote the complete \(K\times1\) regressor vector, where \(K=2k+s+3\). Bold \(\boldsymbol\alpha_*\) is the corresponding \(K\times1\) vector of original coefficients; scalar \(\alpha_*\) denotes the intercept alone. Stacking the \(T\) observation rows \(z_t’\) forms a \(T\times K\) design matrix. Its cross-product matrix is \(\sum_t z_tz_t’\), with dimension \(K\times K\). The regressor–error sum \(\sum_t z_te_t\) is \(K\times1\), because each error \(e_t\) is scalar. Multiplying this vector by the \(K\times K\) inverse gives the \(K\times1\) coefficient-error vector in the exact OLS identity:
To use Equation (11) asymptotically, the usual argument would normalize the two expressions on the right and replace the normalized inverse matrix by the inverse of its limit. That step fails here because the limiting matrix has no inverse. Establishing the estimator’s behavior therefore requires retaining the smaller stochastic terms before taking the limit. Cho’s transformations provide a representation in which those terms survive at their appropriate scales.
Main takeaway. The initial normalization leaves several regressor columns carrying only the same deterministic trend, making the limiting matrix singular. The observed regression can still be estimable because the smaller stochastic fluctuations distinguish its columns.
Every matrix entry, its normalizer, and a finite-sample example: Appendix C.
5. What Frisch–Waugh–Lovell accomplishes
The difficulty comes from several regressors carrying the same large trend. To see their separate contributions, Cho rewrites the columns of the regression. The fitted relationship stays the same, but the new columns distinguish the common trend from the smaller fluctuations around it. The Frisch–Waugh–Lovell argument establishes the exact correspondence between the two sets of coefficients.
Start with the lagged outcome, \(y_{t-1}\). Regress it on the partial sums, the time trend, and a constant. This gives a fitted part and a residual:
Here \(\widetilde\beta_T\) is \(2k\times1\), while \(\widetilde\zeta_T\), \(\widetilde\nu_T\), and \(\widetilde u_{t-1}\) are scalars. The auxiliary regression uses \(2k\) partial-sum columns, one trend, and one constant, so its design matrix is \(T\times d\), with \(d=2k+2\). The fitted part records what those regressors jointly explain in the observed sample. The residual records what remains. Under the maintained cointegrating model, it estimates the deviation from the long-run relationship. It has the scale of a stationary variable, even though the variables used to construct it contain trends.
Now replace the column \(y_{t-1}\) in the dynamic regression by the column \(\widetilde u_{t-1}\). Keep the partial sums, trend, constant, and every short-run regressor. Nothing needed to reconstruct the old column has been discarded: its fitted part is made entirely from columns still present. Conversely, the displayed identity reconstructs the residual from the original columns. The two regressions can therefore produce exactly the same fitted values, with their coefficients expressed differently.
Substituting into the fitted equation. Let \(\widehat{\Delta y_t}\) denote the value fitted by the original dynamic OLS regression. Substitute the auxiliary decomposition for its lagged-outcome column:
\[ \begin{aligned} \widehat{\Delta y_t} &=\widehat\rho_Ty_{t-1}+\widehat\theta_T’\ddot x_{t-1} +\widehat\xi_Tq_t+\widehat\alpha_T+\widehat\kappa_T’z_{2t}\\ &=\widehat\rho_T\left[ \widetilde\beta_T’\ddot x_{t-1}+\widetilde\zeta_Tq_t +\widetilde\nu_T+\widetilde u_{t-1}\right]\\ &\quad+\widehat\theta_T’\ddot x_{t-1} +\widehat\xi_Tq_t+\widehat\alpha_T+\widehat\kappa_T’z_{2t}\\ &=\widehat\rho_T\widetilde u_{t-1} +\left(\widehat\theta_T+\widehat\rho_T\widetilde\beta_T\right)’\ddot x_{t-1}\\ &\quad+\left(\widehat\xi_T+\widehat\rho_T\widetilde\zeta_T\right)q_t +\left(\widehat\alpha_T+\widehat\rho_T\widetilde\nu_T\right) +\widehat\kappa_T’z_{2t}. \end{aligned} \]For example, the partial-sum column receives one contribution from its original coefficient and another from the fitted part of \(y_{t-1}\). Adding them gives \(\widehat\theta_T+\widehat\rho_T\widetilde\beta_T\). The same grouping explains the trend and intercept corrections. The last display fits exactly the same number at every date. The replacement can also be reversed, so the two regressions span the same fitted values and yield the same OLS residuals.
Keeping all the controls matters. The auxiliary residual is orthogonal to the partial sums, trend, and constant used to construct it. It need not be orthogonal to the short-run regressors. Those variables must remain in the dynamic regression and be estimated jointly. Although the proof uses successive regressions, this augmented representation reproduces the original one-step OLS estimator. It is distinct from adopting the separate two-step NARDL estimation procedure discussed later.
The partial sums themselves still carry large deterministic drifts. Cho therefore projects each partial-sum column on the time column \(q_t=t-1\), and subtracts the fitted line:
The numerator in \(\ddot\mu_T\) is a \(2k\times1\) vector, and its denominator is the scalar sum of squared trend values. Thus \(\ddot\mu_T\) is \(2k\times1\), with one estimated drift slope per partial-sum series. The residual \(\ddot m_{t-1}\) is also \(2k\times1\). It retains the cumulative stochastic fluctuations around those fitted drifts. Removing a deterministic line does not remove the accumulated random shocks: these detrended partial sums still have integrated stochastic variation.
This particular projection uses the trend alone, so its fitted line goes through the origin. A constant is retained separately in the subsequent regressions. The resulting condition is \(\sum_tq_t\ddot m_{t-1}=0\), which says that the residual has zero sample cross-product with the trend. It does not say that its sample mean is zero. Demeaning would subtract an average; this operation subtracts a different fitted value, \(\ddot\mu_Tq_t\), at each date. Including an intercept in this projection would require different coefficient bookkeeping.
A three-observation calculation. Consider the illustrative positive partial-sum column \((0,1,3)’\) at dates with trend values \((0,1,2)’\). Its fitted drift is
\[ \ddot\mu_T^+ =\frac{0(0)+1(1)+2(3)}{0^2+1^2+2^2} =\frac{7}{5}. \]Multiply this slope by each trend value and subtract the resulting fitted column:
\[ \begin{pmatrix}\ddot m_0^+\\\ddot m_1^+\\\ddot m_2^+\end{pmatrix} = \begin{pmatrix}0\\1\\3\end{pmatrix} -\frac75\begin{pmatrix}0\\1\\2\end{pmatrix} = \begin{pmatrix}0-0\\1-7/5\\3-14/5\end{pmatrix} = \begin{pmatrix}0\\-2/5\\1/5\end{pmatrix}. \]Now multiply the trend row by the residual column:
\[ \begin{pmatrix}0&1&2\end{pmatrix} \begin{pmatrix}0\\-2/5\\1/5\end{pmatrix} =0(0)+1(-2/5)+2(1/5)=0. \]This is the promised orthogonality. The residual mean is nevertheless \([0-2/5+1/5]/3=-1/15\). This arithmetic example shows what projecting on the trend alone accomplishes.
The transformed regression contains an estimated stationary equilibrium error, detrended cumulative processes, and a deterministic trend. The table gives the column rescalings used in the proof. Each divisor is chosen to keep a column’s sum of squares on a finite scale; it is not yet a convergence rate for an original coefficient. The calculations below explain each row.
| Regressor after transformation | Order of the sum of squared entries | Divide every entry by |
|---|---|---|
| Estimated cointegration residual | \(T\) | \(\sqrt T\) |
| Detrended partial sum | \(T^2\) | \(T\) |
| Time trend | \(T^3\) | \(T^{3/2}\) |
| Constant and stationary controls | \(T\) | \(\sqrt T\) |
Table notes. Here \(T\) is the number of observations used in the regression. A “column” is the full sequence of observations for one regressor. For example, the time-trend column is \((0,1,\ldots,T-1)’\). For a generic scalar regressor \(w_t\), its sum of squared entries is \(\sum_{t=1}^{T}w_t^2\). This sum occupies that regressor’s diagonal position in the OLS cross-product matrix.
The middle column reports how this sum grows as observations are added. An entry \(T^2\), for example, says that the sum has growth order \(T^2\); it does not say that its observed value is exactly \(T^2\). Fixed scale factors and random variation remain in the normalized quantity.
Why the final column takes a square root. If each observation \(w_t\) is divided by a divisor \(\lambda_T\), the sum of squares is divided by \(\lambda_T^2\):
\[ \begin{aligned} \begin{pmatrix}w_1/\lambda_T&\cdots&w_T/\lambda_T\end{pmatrix} \begin{pmatrix}w_1/\lambda_T\\\vdots\\w_T/\lambda_T\end{pmatrix} &=\frac{w_1^2}{\lambda_T^2}+\cdots+\frac{w_T^2}{\lambda_T^2}\\ &=\frac{1}{\lambda_T^2}\sum_{t=1}^{T}w_t^2. \end{aligned} \]We choose \(\lambda_T\) so that the denominator matches the growing power in the sum of squares. A sum growing like \(T\) calls for \(\lambda_T=\sqrt T\); one growing like \(T^2\) calls for \(\lambda_T=T\); and one growing like \(T^3\) calls for \(\lambda_T=T^{3/2}\). This explains the relationship between the last two columns of the table.
Estimated cointegration residual. The first auxiliary regression subtracts the fitted long-run relationship from \(y_{t-1}\). Its residual \(\widetilde u_{t-1}\) estimates the stationary deviation from equilibrium. Under the paper’s results, the generated residual has a typical magnitude that remains on a fixed scale as \(T\) increases. Squaring preserves that fixed scale; adding \(T\) squared observations gives a sum of order \(T\). Dividing every residual by \(\sqrt T\) therefore divides that sum by \(T\).
Detrended partial sum. Subtracting a fitted deterministic line from a partial sum leaves its accumulated random fluctuations. Each component of \(\ddot m_{t-1}\) still has stochastic magnitude \(\sqrt T\) at times proportional to \(T\). Its square has magnitude \(T\), and adding the squared observations produces a sum of order \(T^2\). This gives the divisor \(T\). Detrending and forming the cointegration residual have different effects: the former removes a time pattern, while the latter combines cointegrated variables to recover their stationary equilibrium error.
The divisor must account for all the observations being added. If we divided a detrended partial-sum column by only \(\sqrt T\), its sum of squares would be divided by \(T\). Since the original sum has order \(T^2\), the result would still grow like \(T\). Dividing the column by \(T\) divides its sum of squares by \(T^2\), giving the finite stochastic scale needed in the limit calculation.
Time trend. The trend value \(q_t=t-1\) has magnitude proportional to \(T\) at a fixed fraction of the sample. Squaring gives magnitude \(T^2\), and the exact sum grows like \(T^3/3\), as shown in Section 4. The column divisor is therefore \(\sqrt{T^3}=T^{3/2}\). Its normalized sum of squares tends to \(1/3\).
Constant and stationary controls. The constant column contains \(T\) ones, whose squares add up to exactly \(T\). Dividing every one by \(\sqrt T\) makes the sum of squares exactly one. Under the maintained moment and law-of-large-numbers conditions, a stationary control also has a sum of squares growing in proportion to \(T\), giving the same divisor. Its observations can take many values; their typical magnitude remains on a fixed scale as the sample grows.
A numerical reading of the final column. With \(T=100\), divide each estimated cointegration residual by \(10\), each detrended partial-sum observation by \(100\), each trend observation by \(1000\), and each constant or stationary-control observation by \(10\). These are the theoretical column rescalings used to analyze the OLS matrix. Direct OLS can still be computed from the original model.
After scaling, the columns’ sums of squares remain on finite scales. The normalized sums involving cumulative processes generally have random limits. Lemma 2 establishes that the full transformed cross-product matrix has a nonsingular limit, allowing its inverse to enter the asymptotic argument. The original coefficient estimates’ convergence rates also depend on the error terms and the coefficient transformations developed next.
Main takeaway. The FWL transformations preserve the fitted regression while separating the equilibrium error, cumulative fluctuations, and time trend. Giving each type of column its own normalization retains the variation needed for a nonsingular limiting matrix.
Exact FWL proof: Appendix D. Detrending and coefficient adjustments: Appendix E.
6. Recovering every original coefficient
After detrending, the auxiliary long-run regression has the exact fitted-value decomposition
\[ y_{t-1}=\ddot\beta_T’\ddot m_{t-1}+\ddot\vartheta_Tq_t+\ddot\nu_T+\ddot u_{t-1}. \]The vectors \(\ddot\beta_T\) and \(\ddot m_{t-1}\) are \(2k\times1\); the trend coefficient \(\ddot\vartheta_T\), intercept \(\ddot\nu_T\), and residual \(\ddot u_{t-1}\) are scalars. The long-run regression’s residual from Equation (12) is unchanged: \(\ddot u_{t-1}=\widetilde u_{t-1}\). Its slope coefficients and intercept are also unchanged. Its trend coefficient absorbs the fitted trends moved out of the partial sums; the correction \(\ddot\mu_T’\ddot\beta_T\) is a scalar product of a \(1\times2k\) row and a \(2k\times1\) column:
The transformed dynamic regression is then
In this dynamic regression, \(\ddot\eta_T\) is a \(2k\times1\) coefficient vector and \(\ddot\kappa_T\) is \(s\times1\). The coefficients \(\ddot\rho_T\), \(\ddot\varsigma_T\), and \(\ddot\gamma_T\) are scalars, as are the residuals \(\ddot u_{t-1}\) and \(\ddot e_t\). There are still \(K\) regressor columns, giving a \(T\times K\) design matrix. The scalar coefficients \(\ddot\vartheta_T\) and \(\ddot\varsigma_T\) belong to different equations: the first describes the trend in the fitted long-run relationship; the second describes the trend in this dynamic regression.
Undo the two substitutions. Express both transformed regressors in the original columns:
\[ \begin{aligned} \ddot u_{t-1} &=y_{t-1}-\ddot\beta_T’\ddot x_{t-1} -\widetilde\zeta_Tq_t-\ddot\nu_T,\\ \ddot m_{t-1}&=\ddot x_{t-1}-\ddot\mu_Tq_t. \end{aligned} \]The first line uses Equation (14), the unchanged auxiliary residual, and the original trend coefficient \(\widetilde\zeta_T\). Inserting these two expressions into Equation (15) gives
\[ \begin{aligned} \Delta y_t={}&\ddot\rho_T \left[y_{t-1}-\ddot\beta_T’\ddot x_{t-1} -\widetilde\zeta_Tq_t-\ddot\nu_T\right]\\ &+\ddot\eta_T’\left[\ddot x_{t-1}-\ddot\mu_Tq_t\right] +\ddot\varsigma_Tq_t+\ddot\gamma_T +\ddot\kappa_T’z_{2t}+\ddot e_t\\ ={}&\ddot\rho_Ty_{t-1} +\left(\ddot\eta_T-\ddot\rho_T\ddot\beta_T\right)’\ddot x_{t-1}\\ &+\left(\ddot\varsigma_T-\ddot\mu_T’\ddot\eta_T -\ddot\rho_T\widetilde\zeta_T\right)q_t\\ &+\left(\ddot\gamma_T-\ddot\rho_T\ddot\nu_T\right) +\ddot\kappa_T’z_{2t}+\ddot e_t. \end{aligned} \]Read the coefficient attached to each original regressor in the last expression and compare it with Equation (3). This yields the reconstruction identities of Proposition 4:
Consider one coordinate \(j\) of the partial-sum vector. The estimated disequilibrium contains the term \(-\ddot\beta_{T,j}\ddot x_{t-1,j}\). Multiplying disequilibrium by \(\ddot\rho_T\) therefore contributes \(-\ddot\rho_T\ddot\beta_{T,j}\) to the coefficient on that partial sum. The transformed dynamic equation also contributes its direct coefficient \(\ddot\eta_{T,j}\). Adding both contributions gives \(\widehat\theta_{T,j}\).
For example, suppose the estimates are \(\ddot\beta_{T,j}=2\), \(\ddot\rho_T=-0.4\), and \(\ddot\eta_{T,j}=0.03\). Then
\[ \widehat\theta_{T,j}=0.03-(-0.4)(2)=0.03+0.80=0.83. \]The contribution through estimated disequilibrium is \(0.80\); the additional partial-sum contribution is \(0.03\). The value \(2\) is an auxiliary estimate of a long-run slope. The value \(0.83\) is the reconstructed coefficient in the original dynamic equation. At the population level, the corresponding parameters satisfy \(\theta_{*,j}=-\rho_*\beta_{*,j}\).
The long-run slope reported from the one-step regression is its coefficient ratio:
\[ -\frac{\widehat\theta_{T,j}}{\widehat\rho_T} =-\frac{0.83}{-0.4}=2.075. \]The auxiliary regression gives \(2\); the one-step ratio gives \(2.075\). They estimate the same population slope through different finite-sample calculations. The exact reconstruction reproduces the original dynamic OLS coefficients; it does not make these two long-run estimates identical.
The trend row has two corrections because both transformations moved terms into the time column. Subtracting \(\ddot\mu_T’\ddot\eta_T\) reverses the drift removed from the partial sums. Subtracting \(\ddot\rho_T\widetilde\zeta_T\) reverses the explicit trend removed when constructing disequilibrium. The intercept similarly subtracts \(\ddot\rho_T\ddot\nu_T\). The lagged-outcome and short-run coefficients require no adjustment.
This example also explains why knowing the precision of the auxiliary long-run slope is insufficient. The reconstructed coefficient depends on the adjustment estimate and the auxiliary dynamic coefficient as well. An error in any of these estimated quantities can change the final value. The next section follows these separate errors and determines which contributions remain after scaling by the sample size.
Write the reconstruction as a matrix multiplication. Stack the original and transformed estimates in the same regressor order:
\[ \widehat{\boldsymbol\alpha}_T= \begin{pmatrix} \widehat\rho_T\\ \widehat\theta_T\\ \widehat\xi_T\\ \widehat\alpha_T\\ \widehat\kappa_T \end{pmatrix}, \qquad \ddot\tau_T= \begin{pmatrix} \ddot\rho_T\\ \ddot\eta_T\\ \ddot\varsigma_T\\ \ddot\gamma_T\\ \ddot\kappa_T \end{pmatrix}. \]These are block columns. Each contains one adjustment coefficient, \(2k\) partial-sum coefficients, one trend coefficient, one intercept, and \(s\) short-run coefficients. Thus each has \(1+2k+1+1+s=K\) rows and one column. Bold \(\widehat{\boldsymbol\alpha}_T\) denotes the whole coefficient vector; the unbolded \(\widehat\alpha_T\) in its fourth block denotes the scalar intercept.
The matrix that performs all the operations in Equation (16) is
\[ \mathbf T_T= \begin{pmatrix} 1&0&0&0&0\\ -\ddot\beta_T&I_{2k}&0&0&0\\ -\widetilde\zeta_T&-\ddot\mu_T’&1&0&0\\ -\ddot\nu_T&0&0&1&0\\ 0&0&0&0&I_s \end{pmatrix}, \qquad \widehat{\boldsymbol\alpha}_T=\mathbf T_T\ddot\tau_T. \]The block row heights and column widths are \(1,2k,1,1,s\), so \(\mathbf T_T\) is \(K\times K\). In the second block row, \(-\ddot\beta_T\) is a \(2k\times1\) column: multiplying it by the scalar \(\ddot\rho_T\), then adding \(I_{2k}\ddot\eta_T\), produces the \(2k\times1\) vector \(\widehat\theta_T\). In the third row, \(-\ddot\mu_T’\) is a \(1\times2k\) row. Multiplication by the \(2k\times1\) column \(\ddot\eta_T\) produces the scalar drift correction:
\[ (-\ddot\mu_T’)\ddot\eta_T =-\sum_{j=1}^{2k}\ddot\mu_{T,j}\ddot\eta_{T,j}. \]That third row also multiplies \(-\widetilde\zeta_T\) by \(\ddot\rho_T\), then adds \(\ddot\varsigma_T\). It therefore restores both contributions to the original time trend. The fourth row restores the intercept by subtracting \(\ddot\rho_T\ddot\nu_T\). The first and last rows copy the adjustment and short-run coefficients. Every zero in the display is a block whose dimensions are fixed by its row height and column width; for example, the last zero in the first row is \(1\times s\).
With full sample rank, this matrix equality is exact. Equation (E.3) derives \(\mathbf T_T=\mathbf R_T\mathbf Q_T\) by combining the two transformations, applied from right to left. The bold symbol \(\mathbf T_T\) names a matrix; the subscript \(T\) records its dependence on the sample size. Both this matrix and the transformed estimates contain estimated quantities. Their joint uncertainty must therefore be carried through the multiplication to obtain the limiting behavior of the original OLS estimator.
Main takeaway. Every original OLS coefficient can be recovered exactly from the transformed coefficients and the auxiliary estimates. These identities allow the probability limits derived in the transformed regression to describe the estimator used in applied work.
Row-by-row reconstruction, including the trend and intercept: Appendix E.
7. From exact algebra to probability limits
Equation (16) identifies how the original OLS estimates depend on the auxiliary estimates. Lemmas 3 and 4 establish how their estimation errors behave as the sample grows. Their central concern is that the transformed regressors themselves contain estimated quantities: the fitted drift enters the detrended partial sums, and the estimated long-run coefficients enter the cointegration residual.
Consider first the long-run regression. Begin with its population relationship and replace the partial sums by their fitted trends plus detrended values:
\[ \begin{aligned} y_{t-1} &=\beta_*’\ddot x_{t-1}+\zeta_*q_t+\nu_*+u_{t-1}\\ &=\beta_*’\left(\ddot m_{t-1}+q_t\ddot\mu_T\right) +\zeta_*q_t+\nu_*+u_{t-1}\\ &=\beta_*’\ddot m_{t-1} +\underbrace{\left(\zeta_*+\ddot\mu_T’\beta_*\right)}_{\vartheta_{T*}}q_t +\nu_*+u_{t-1}. \end{aligned} \]The coefficient required on the transformed trend is therefore \(\vartheta_{T*}=\zeta_*+\ddot\mu_T’\beta_*\). It depends on the sample because the fitted drift defines the transformed regressors. A different estimated drift changes the transformed trend coefficient by the corresponding long-run-slope-weighted amount. The underlying population slopes \(\beta_*\) and trend coefficient \(\zeta_*\) remain fixed.
The same relationship written with two fitted drifts. Take fixed long-run slopes \(\beta_*^+=2\), \(\beta_*^-=1\), and fixed trend coefficient \(\zeta_*=0.1\). The partial-sum and trend contribution to the original relationship is
\[ 2x_{t-1}^++x_{t-1}^-+0.1q_t. \]First suppose the fitted drifts are \(0.75\) and \(-0.75\). Subtracting those fitted trends gives \(m_{t-1}^{+,A}=x_{t-1}^+-0.75q_t\) and \(m_{t-1}^{-,A}=x_{t-1}^-+0.75q_t\). The superscript \(A\) labels this illustrative choice. The two drift choices in this example illustrate possible reparameterizations; a given sample and projection rule produce a single fitted drift. The transformed trend coefficient must be
\[ \vartheta_A=0.1+2(0.75)+1(-0.75)=0.85. \]Check the whole contribution by putting the definitions back in:
\[ \begin{aligned} 2m_{t-1}^{+,A}+m_{t-1}^{-,A}+0.85q_t &=2(x_{t-1}^+-0.75q_t) +(x_{t-1}^-+0.75q_t)+0.85q_t\\ &=2x_{t-1}^++x_{t-1}^-+(-1.5+0.75+0.85)q_t\\ &=2x_{t-1}^++x_{t-1}^-+0.1q_t. \end{aligned} \]Now use fitted drifts \(0.80\) and \(-0.70\), labeled \(B\). The detrended values become \(m_{t-1}^{+,B}=x_{t-1}^+-0.80q_t\) and \(m_{t-1}^{-,B}=x_{t-1}^-+0.70q_t\). The required trend coefficient becomes \(\vartheta_B=0.1+2(0.80)+1(-0.70)=1.00\), and
\[ \begin{aligned} 2m_{t-1}^{+,B}+m_{t-1}^{-,B}+1.00q_t &=2x_{t-1}^++x_{t-1}^-+(-1.60+0.70+1.00)q_t\\ &=2x_{t-1}^++x_{t-1}^-+0.1q_t. \end{aligned} \]Both expressions recover the same original contribution. The economic slopes \(2\) and \(1\), and the original trend coefficient \(0.1\), remain fixed. What changes is the allocation of the time pattern between the detrended columns and the explicit trend. Because an actual fitted drift depends on the sample, the coefficient required by that particular transformation is written \(\vartheta_{T*}\).
Why a shrinking error can have a random limit
Before reading Lemma 3, distinguish two questions. Convergence in probability to a fixed target asks whether an error larger than any fixed tolerance becomes unlikely. Convergence in distribution, denoted \(\Rightarrow\), asks whether the probabilities describing a random quantity approach those of a limiting random variable. For a scalar quantity \(Z_T\) and limiting variable \(Z\), its definition is
The continuity qualification in Equation (17) excludes thresholds at which the limiting distribution places positive probability on a single value. For a continuous normal distribution, every threshold qualifies. Imagine a histogram made from many possible samples: convergence in distribution describes the limiting shape of that histogram. It does not say that each realized value approaches one particular random draw.
The two questions work together when we multiply a shrinking estimation error by a growing factor. In the earlier teaching example, \(X_T=U/\sqrt T\) converges to zero in probability, but \(\sqrt T X_T=U\) has the same standard normal distribution at every sample size. Multiplying by \(\sqrt T\) keeps the shrinking uncertainty visible. The paper determines the appropriate multiplier and limiting distribution for each of its estimates.
Lemma 3 describes the errors around the targets above. Let \(\mathcal L_1\) denote the \(2k\times1\) limiting slope-error vector, \(\mathcal L_2\) the scalar limiting transformed-trend error, and \(\mathcal L_3\) the scalar limiting intercept error. Stacking them gives a \(d\times1\) joint random vector \(\mathcal L\), with \(d=2k+2\). The display below has three blocks, and its first block contains \(2k\) entries. Both sides therefore have \(d\) entries. The distributional statement is joint: it preserves the dependence among these scaled errors as well as their individual distributions.
Each multiplier indicates the scale needed to preserve a nonzero fluctuation. The slope error is of order \(T^{-1}\); the intercept error is of order \(T^{-1/2}\). The transformed-trend error is of order \(T^{-3/2}\) around its sample-dependent target. The unscaled errors approach zero, establishing consistency, while their scaled limits describe the remaining uncertainty.
The symbols \(\mathcal L_j\) describe the limiting distributions of errors across samples generated by the model. For the long-run slope, two statements work together:
\[ \ddot\beta_T\xrightarrow{p}\beta_*, \qquad T(\ddot\beta_T-\beta_*)\Rightarrow\mathcal L_1. \]The first statement concerns the shrinking error itself; the second concerns that error multiplied by \(T\). For one slope component, convergence in probability means that the chance of an error exceeding, say, \(0.01\) tends to zero—and the same is true for any smaller fixed tolerance. Its distributional limit describes probabilities at a tolerance that shrinks with the sample. For example,
\[ \Pr\!\left(|\ddot\beta_{T,j}-\beta_{*,j}|\leq\frac{2}{T}\right) =\Pr\!\left(|T(\ddot\beta_{T,j}-\beta_{*,j})|\leq2\right) \longrightarrow\Pr\!\left(|\mathcal L_{1,j}|\leq2\right), \]provided the limiting distribution has no probability mass at the endpoints \(-2\) and \(2\). The left-hand interval narrows as \(T\) grows. The chance that the estimation error lies inside this shrinking interval need not approach one: the scaled distribution describes how much uncertainty remains at that increasingly fine scale. This is how consistency and a random limiting distribution fit together.
Centering the transformed trend coefficient. The middle row of Equation (18) compares the estimate with the coefficient required by the fitted drift in that sample. Write this target beside its fixed limiting value:
\[ \vartheta_{T*}=\zeta_*+\ddot\mu_T’\beta_*, \qquad \vartheta_*:=\zeta_*+\mu_*’\beta_*. \]The vector \(\ddot\mu_T\) contains the estimated drifts, while \(\mu_*\) contains their population values. Both are \(2k\times1\), as is the fixed long-run slope vector \(\beta_*\). The subscript \(T\) on \(\vartheta_{T*}\) records its dependence on the sample. Its asterisk does not make it a fixed quantity: it combines the fixed parameters with the estimated drift. Since \(\ddot\mu_T\xrightarrow{p}\mu_*\), the target \(\vartheta_{T*}\) converges in probability to \(\vartheta_*\).
Subtract the definitions, cancel the two copies of \(\zeta_*\), and collect the drift difference:
\[ \begin{aligned} \vartheta_{T*}-\vartheta_* &=(\zeta_*+\ddot\mu_T’\beta_*)-(\zeta_*+\mu_*’\beta_*)\\ &=\zeta_*+\ddot\mu_T’\beta_*-\zeta_*-\mu_*’\beta_*\\ &=(\ddot\mu_T-\mu_*)’\beta_*\\ &=\beta_*'(\ddot\mu_T-\mu_*). \end{aligned} \]The final equality reverses the order of a scalar inner product. Both expressions add the same \(2k\) products, \(\sum_{j=1}^{2k}\beta_{*,j}(\ddot\mu_{T,j}-\mu_{*,j})\). A \(1\times2k\) row multiplied by a \(2k\times1\) column leaves a single number.
Connect this calculation to the earlier example. Take \(\beta_*=(2,1)’\), \(\zeta_*=0.1\), population drifts \(\mu_*=(0.75,-0.75)’\), and fitted drifts \(\ddot\mu_T=(0.80,-0.70)’\). The fixed target is \(0.85\), while the sample-dependent target is \(1.00\). Their difference is exactly
\[ \begin{aligned} &\begin{pmatrix}2&1\end{pmatrix} \left[ \begin{pmatrix}0.80\\-0.70\end{pmatrix} -\begin{pmatrix}0.75\\-0.75\end{pmatrix} \right]\\ &\qquad=\begin{pmatrix}2&1\end{pmatrix} \begin{pmatrix}0.05\\0.05\end{pmatrix} =2(0.05)+1(0.05)=0.15. \end{aligned} \]The change in the fitted drift reallocates the time pattern between the detrended regressors and the explicit trend. The original population relationship remains the same.
To compare the estimate \(\ddot\vartheta_T\) with the fixed target, add and subtract \(\vartheta_{T*}\). There are then two sources of error:
\[ \begin{aligned} \ddot\vartheta_T-\vartheta_* &=(\ddot\vartheta_T-\vartheta_{T*})+(\vartheta_{T*}-\vartheta_*)\\ &=(\ddot\vartheta_T-\vartheta_{T*}) +\beta_*'(\ddot\mu_T-\mu_*). \end{aligned} \]The first term measures the precision of the transformed regression around its own target; Lemma 3 gives it order \(O_p(T^{-3/2})\). The second measures the shift in that target caused by estimating the drift. It is \(O_p(T^{-1/2})\) and can be the larger contribution. Multiplying the whole expression by \(T^{3/2}\) also multiplies this target shift. Its convergence to zero on the original scale therefore does not justify dropping it from the scaled calculation. This is why the middle row of Equation (18) keeps \(\vartheta_{T*}\) as its centering value.
Returning to the original trend coefficient makes the role of slope estimation explicit. Proposition 2 gives the following exact cancellation:
\[ \begin{aligned} \widetilde\zeta_T-\zeta_* &=\ddot\vartheta_T-\ddot\mu_T’\ddot\beta_T-\zeta_*\\ &=(\ddot\vartheta_T-\vartheta_{T*}) +\left(\vartheta_{T*}-\zeta_*\right) -\ddot\mu_T’\ddot\beta_T\\ &=(\ddot\vartheta_T-\vartheta_{T*}) +\ddot\mu_T’\beta_*-\ddot\mu_T’\ddot\beta_T\\ &=(\ddot\vartheta_T-\vartheta_{T*}) -\ddot\mu_T'(\ddot\beta_T-\beta_*). \end{aligned} \]The first term has order \(T^{-3/2}\), whereas the second has order \(T^{-1}\). The original explicit-trend estimate therefore inherits the larger slope-error contribution. Multiplying by \(T\) removes the first term in the limit and gives \(T(\widetilde\zeta_T-\zeta_*)\Rightarrow-\mu_*’\mathcal L_1\). This explains why the original and transformed trend coefficients require different normalizations.
Lemma 4 then follows the long-run estimation errors into the dynamic regression. Write those errors as \(b_T=\ddot\beta_T-\beta_*\), \(v_T=\ddot\vartheta_T-\vartheta_{T*}\), and \(n_T=\ddot\nu_T-\nu_*\). The slope-error vector \(b_T\) is \(2k\times1\); the trend and intercept errors \(v_T\) and \(n_T\) are scalars. This scalar \(v_T\) is distinct from the \(d\times1\) observation vector \(v_t\) in Section 4. The product \(b_T’\ddot m_{t-1}\) is scalar, so every term in the following residual identity is scalar:
\[ \begin{aligned} u_{t-1}-\ddot u_{t-1} &=\left[y_{t-1}-\beta_*’\ddot m_{t-1}-\vartheta_{T*}q_t-\nu_*\right]\\ &\quad-\left[y_{t-1}-\ddot\beta_T’\ddot m_{t-1}-\ddot\vartheta_Tq_t-\ddot\nu_T\right]\\ &=(\ddot\beta_T-\beta_*)’\ddot m_{t-1} +(\ddot\vartheta_T-\vartheta_{T*})q_t+(\ddot\nu_T-\nu_*)\\ &=b_T’\ddot m_{t-1}+v_Tq_t+n_T. \end{aligned} \]The two copies of \(y_{t-1}\) cancel. Adding \(\ddot u_{t-1}\) to both sides gives
For example, overestimating a long-run slope subtracts too much of its regressor when constructing \(\ddot u_{t-1}\). The retained regressor must compensate for that error in the dynamic equation. Substituting the identity above into \(\Delta y_t=\rho_*u_{t-1}+\gamma_*+\kappa_*’z_{2t}+e_t\) produces
\[ \Delta y_t=\rho_*\ddot u_{t-1} +\rho_*b_T’\ddot m_{t-1} +\rho_*v_Tq_t+\gamma_*+\rho_*n_T +\kappa_*’z_{2t}+e_t. \]The appropriate sample-dependent targets are consequently \(\rho_*b_T\) for the retained level coefficients, \(\rho_*v_T\) for the trend, and \(\gamma_*+\rho_*n_T\) for the intercept. Their first-stage errors shrink, but they remain relevant at the scales used to derive the distributions.
As a scalar illustration, suppose the slope error is \(b_T=0.02\) and \(\rho_*=-0.5\). The generated residual subtracts an extra \(0.02\) times the detrended regressor. Multiplication by the adjustment coefficient introduces an extra positive contribution of \(0.01\) times that regressor. The coefficient \(\rho_*b_T=-0.01\) on the retained level offsets it. This compensation is the reason for the centering term in Lemma 4.
Two blocks of Lemma 4 are especially useful for understanding the main result. Denote the scalar limiting adjustment-coefficient error by \(\mathcal D_1\) and the \(2k\times1\) limiting centered level-coefficient error vector by \(\mathcal D_2\). They are the first two blocks of the full \(K\times1\) dynamic-limit vector \(\mathcal D\). Its remaining blocks concern the scalar trend coefficient, scalar intercept, and \(s\times1\) short-run coefficient vector. Then
The second limit in Equation (20) centers the estimate at the coefficient required after substituting Equation (19) into the dynamic equation. Centering directly at zero would retain the additional first-stage term. The same reasoning governs the transformed trend and intercept. The limiting quantities are jointly determined by the observations, so their dependence must be preserved when reconstructing OLS.
After this centering, the primitive dynamic errors have a mixed-normal limit. To read this statement, imagine holding the limiting regressor information fixed: conditional on that information, the error distribution is Gaussian. Different realizations of the regressors can produce different covariance matrices. The term “mixed normal” describes this Gaussian distribution with a random covariance. Lemma 3’s static long-run errors generally have a different, nonstandard limit. The next step shows how the reconstruction removes their contribution from the leading distribution of the original estimator.
Both lemmas obtain their limits from normalized regressor cross-products and regressor–error products, often called scores. The transformed regressions retain the accumulated fluctuations at their appropriate scale, so their limiting moment matrices can be inverted. Combining these inverse matrices with the corresponding score limits gives the joint coefficient distributions. The Brownian expressions in the paper specify those random limits; the centering identities above explain which estimation errors they describe.
Main takeaway. Estimating the long-run relationship also changes the regressors used in the dynamic equation. The proof follows those first-stage errors jointly with the dynamic-regression errors and keeps the appropriate centering explicit.
Lemma 3 and its centering: Appendix F. Lemma 4 and generated-regressor uncertainty: Appendix G.
8. Why the convergence rate depends on the coefficient
The equation-level coefficient combines the estimated equilibrium slope with the estimated speed of adjustment. The reconstruction in Equation (16) gives \(\widehat\theta_T=\ddot\eta_T-\ddot\rho_T\ddot\beta_T\), while its population value is \(\theta_*=-\rho_*\beta_*\). Write \(a_T:=\ddot\rho_T-\rho_*\) for the scalar adjustment error, and retain \(b_T:=\ddot\beta_T-\beta_*\) for the slope-error vector. Thus \(\ddot\rho_T=\rho_*+a_T\) and \(\ddot\beta_T=\beta_*+b_T\).
Substitute these two expressions, distribute the product, and cancel the population term:
\[ \begin{aligned} \widehat\theta_T-\theta_* &=\ddot\eta_T-(\rho_*+a_T)(\beta_*+b_T)+\rho_*\beta_*\\ &=\ddot\eta_T-\rho_*\beta_*-\rho_*b_T-a_T\beta_*-a_Tb_T +\rho_*\beta_*\\ &=-\beta_*a_T+\underbrace{(\ddot\eta_T-\rho_*b_T)}_{h_T}-a_Tb_T. \end{aligned} \]Here \(h_T\) names the centered dynamic-coefficient error studied in Lemma 4. It is \(2k\times1\), as is \(b_T\). Writing the same identity without the shorthand gives
Each term in Equation (21) is a \(2k\times1\) vector. The adjustment-speed error \(\ddot\rho_T-\rho_*\) is scalar, so it multiplies every component of the slope vector. The first term is adjustment-speed error multiplied by the true long-run slope. The second is precisely the centered error characterized by Lemma 4. The third is the product of the scalar adjustment error and the \(2k\times1\) slope error. This expansion is exact, including the product term.
When the corresponding component of \(\beta_*\) is nonzero, the first term has order \(T^{-1/2}\). The second has order \(T^{-1}\), and the product has order \(T^{-3/2}\), because \(T^{-1/2}T^{-1}=T^{-3/2}\). The following arithmetic illustrates how their scales separate as the sample grows; the actual errors also contain random multipliers.
| Sample size | Adjustment contribution \(-\beta_*a_T\): scale \(T^{-1/2}\) | Centered dynamic error \(h_T\): scale \(T^{-1}\) | Product of errors \(-a_Tb_T\): scale \(T^{-3/2}\) |
|---|---|---|---|
| \(T=100\) | 0.1 | 0.01 | 0.001 |
| \(T=400\) | 0.05 | 0.0025 | 0.000125 |
Table notes. These entries are arithmetic values of three powers of \(T\), illustrating the scales of the three terms in the preceding expansion. For example, at \(T=400\), \(T^{-1/2}=1/\sqrt{400}=1/20=0.05\). Quadrupling \(T\) divides the three scales by \(2\), \(4\), and \(8\), respectively. Actual estimation errors also contain parameter loadings and random factors; the displayed numbers are neither estimated errors nor standard errors.
Multiply all three terms by the same factor. The powers in the table transform as follows:
\[ \begin{aligned} \sqrt T\times T^{-1/2}&=T^{1/2-1/2}=1,\\ \sqrt T\times T^{-1}&=T^{1/2-1}=T^{-1/2},\\ \sqrt T\times T^{-3/2}&=T^{1/2-3/2}=T^{-1}. \end{aligned} \]The first scale remains visible; the other two shrink to zero. To see the corresponding random quantities, apply the multiplication to the exact error identity and group each normalized error:
\[ \begin{aligned} \sqrt T(\widehat\theta_T-\theta_*) &=-\beta_*\sqrt T a_T+\sqrt T h_T-\sqrt T a_Tb_T\\ &=-\beta_*(\sqrt T a_T) +\frac{Th_T}{\sqrt T} -\frac{(\sqrt T a_T)(Tb_T)}{T}. \end{aligned} \]Equations (18) and (20) establish finite limiting distributions for \(\sqrt T a_T\), \(Th_T\), and \(Tb_T\). These normalized errors are therefore bounded in probability. Dividing the second by \(\sqrt T\), and the product in the third by \(T\), makes those contributions vanish in probability. Independence is unnecessary for this order calculation. Consequently,
Recall the definition of \(o_p(1)\): the omitted difference converges to zero in probability. Here the statement can be written explicitly as
\[ \sqrt T(\widehat\theta_T-\theta_*) +\beta_*\sqrt T(\ddot\rho_T-\rho_*) \xrightarrow{p}0. \]Thus, on the \(\sqrt T\) scale used to study the estimation error, the original level-coefficient error and its leading adjustment-error contribution become arbitrarily close with probability approaching one. The omitted terms remain in the exact finite-sample identity, while their probability of changing this scaled approximation by more than a fixed tolerance tends to zero.
This rate is consistent with the economic parameterization. Holding the equilibrium slopes fixed, a change in the estimated adjustment speed rescales their coefficients in the dynamic equation. A rapidly estimated slope therefore does not make its product with a more slowly estimated adjustment coefficient equally precise.
For an illustrative long-run slope of two, an adjustment-coefficient error of \(0.01\) contributes \(-2(0.01)=-0.02\) to the level-coefficient error. A slope of one receives a contribution of \(-0.01\) from the same adjustment error. These are the loadings in Equation (22). The remaining estimation components affect finite-sample errors but shrink more quickly in the generic case.
The same source determines the leading uncertainty in the original trend coefficient when \(\zeta_*\ne0\). The generic joint limit is
The coefficient-error vector on the left of Equation (23) is \(d\times1\): it contains one adjustment error, \(2k\) level-coefficient errors, and one trend-coefficient error. The loading vector \((1,-\beta_*’,-\zeta_*)’\) on the right has the same dimension and multiplies the single random scalar \(\mathcal D_1\). These leading errors therefore share one random component with different loadings. Their limiting linear dependence is compatible with distinct, computable coefficient estimates in a finite sample: the smaller terms in the expansion still distinguish those estimates.
The intercept also contains a surviving dynamic-regression error. Let the scalar \(\mathcal D_4\) denote the limit of \(\sqrt T\{(\ddot\gamma_T-\gamma_*)-\rho_*n_T\}\), and let the \(s\times1\) vector \(\mathcal D_5\) denote the limit of \(\sqrt T(\ddot\kappa_T-\kappa_*)\). Reconstructing the original intercept cancels its first-stage estimation contribution, giving the scalar intercept limit and the \(s\times1\) short-run coefficient limit:
\[ \sqrt T(\widehat\alpha_T-\alpha_*)\Rightarrow\mathcal D_4-\nu_*\mathcal D_1, \qquad \sqrt T(\widehat\kappa_T-\kappa_*)\Rightarrow\mathcal D_5. \]Together, these expressions describe every original coefficient in the generic case of Theorem 1. Their joint distribution is mixed normal because the surviving components come from the centered dynamic limit. The error terms all vanish without normalization, establishing consistency.
When a population slope is zero. The positive-slope block \(\beta_*^+\) has \(k\) entries, one per original explanatory variable. In the theorem’s case \(\beta_*^+=0\), every entry of this block is zero. Apply that restriction to the same exact expansion:
\[ \widehat\theta_T^+-\theta_*^+ =\underbrace{-\beta_*^+a_T}_{0}+h_T^+-a_Tb_T^+ =h_T^+-a_Tb_T^+. \]The superscript \(+\) selects the positive-partial-sum entries of each vector. The \(T^{-1}\) contribution now leads. Multiplying by \(T\) gives
\[ T(\widehat\theta_T^+-\theta_*^+) =Th_T^+-\frac{(\sqrt T a_T)(Tb_T^+)}{\sqrt T}. \]The first term retains its limit and the second vanishes. Writing \(\mathcal D_2^+\) for the corresponding \(k\times1\) block of \(\mathcal D_2\), the theorem gives
\[ T(\widehat\theta_T^+-\theta_*^+)\Rightarrow\mathcal D_2^+. \]The analogous result holds for a zero negative block. When \(\zeta_*=0\), the trend coefficient likewise requires multiplication by \(T\), with limit \(-\mu_*’\mathcal D_2\). The original \(\sqrt T\) expansion remains valid in these directions, but its limit is zero. A finer normalization reveals the remaining fluctuation. Thus the parameter values determine the informative rate even though the OLS computation itself is unchanged.
Why the long-run ratio converges faster. Define the one-step long-run estimate by \(\widehat\beta_T^{\mathrm{ratio}}=-\widehat\theta_T/\widehat\rho_T\). Use \(\widehat\rho_T=\ddot\rho_T\), the reconstruction formula, and the already defined \(h_T=\ddot\eta_T-\rho_*b_T\):
In Equation (24), the term proportional to the true slope and the adjustment error has canceled. The numerator contains only the smaller product of errors and the centered dynamic error.
The product in the numerator has order \(T^{-3/2}\), and \(h_T\) has order \(T^{-1}\). The ratio therefore has rate \(T\), with scaled limit \(-\mathcal D_2/\rho_*\). This is an algebraic deduction from the preceding identities: the common adjustment-speed error cancels when the equation coefficient is divided by that adjustment estimate. It also explains why convergence rates must be compared for the same target parameter.
Division preserves these orders because \(\ddot\rho_T\) converges to the fixed nonzero \(\rho_*\). Thus the equilibrium slope and its coefficient in the dynamic equation can be consistently estimated at different rates within the same one-step regression.
Main takeaway. The original level coefficients generally inherit the \(\sqrt T\) rate of the adjustment coefficient. When a leading coefficient is zero, a finer scale can reveal the remaining variation; the long-run ratios converge at rate \(T\) because the leading adjustment error cancels.
The complete product expansion, the four cases of Theorem 1, and a long-run ratio calculation: Appendix H.
9. How applied economists should use these results
Use one-step OLS with a clear understanding of its theoretical setting
For an applied economist working with a correctly specified cointegrated NARDL model containing a time trend, Cho supplies a theoretical basis for the familiar one-step OLS estimator. The auxiliary regressions explain that estimator; they are not extra steps that must be run to reproduce its point estimates. They are also distinct from the separate bias-correcting two-stage NARDL procedure.
Report the parameter that answers the economic question
The coefficient \(\theta_*^+\) multiplies the lagged positive partial sum in the dynamic equation. The equilibrium response is \(\beta_*^+=-\theta_*^+/\rho_*\). Reporting these objects under distinct labels helps readers interpret both their economic meaning and their statistical behavior. The adjustment coefficient \(\rho_*\) describes the response to disequilibrium, conditional on the short-run dynamics.
In the one-step regression, the corresponding estimated equilibrium slope is \(-\widehat\theta_T^+/\widehat\rho_T\). The cancellation of the common adjustment-speed error appears in Equation (24) and is developed in Appendix H. The auxiliary long-run estimate \(\ddot\beta_T\) and the one-step ratio are different finite-sample estimators, even when both have rate \(T\).
Compare OLS and two-stage NARDL on the same basis
Cho’s comparison gives a conditional practical recommendation. With a sufficiently large sample, direct OLS can be attractive because it avoids the preliminary long-run covariance estimation used for bias correction in two-stage NARDL. With a smaller sample, two-stage NARDL can be preferable when that correction successfully removes the relevant bias.
| Situation | Practical interpretation |
|---|---|
| A sufficiently large sample within the maintained model | One-step OLS offers a convenient estimator with an established asymptotic justification. |
| A smaller sample in which the two-stage correction works effectively | Bias correction can make two-stage NARDL preferable. |
Table notes. Read each row as a conditional comparison within the maintained cointegrating model. Bias means that the estimator’s average across repeated samples differs from its target. The two-stage correction seeks to remove a systematic component of that error. However, the quantities needed to construct the correction must themselves be estimated. A correction that works effectively can improve small-sample performance; its success cannot be inferred from sample size alone.
The large-sample row explains the appeal of direct OLS: its asymptotic properties are established without requiring the preliminary long-run covariance estimation used by the correction. The table cannot decide which estimator is more accurate for a particular dataset. In particular, it supplies no threshold at which the preferred procedure automatically switches.
A separate issue is comparing convergence rates for the same economic parameter. The \(T\)-rate estimate of the long-run slope \(\beta_*\) and the \(\sqrt T\)-rate estimate of the dynamic coefficient \(\theta_*=-\rho_*\beta_*\) target different quantities. Compare long-run slope estimates with other long-run slope estimates, and adjustment estimates with other adjustment estimates. Consistency describes what happens as the sample grows; finite-sample bias concerns the estimator at the sample size actually available.
Interpret the trend and the matrix problem precisely
The theorem includes an explicit time-trend regressor, even when its true coefficient is zero. That coefficient describes a deterministic time effect conditional on the partial sums and other included regressors. The partial sums themselves already carry drift, so the observed trend in the outcome reflects their contributions as well as the explicit time coefficient.
An asymptotically singular normalized matrix also differs from exact sample collinearity. The former is the theoretical problem resolved here. If software cannot estimate a coefficient because columns are exactly dependent in the observed sample, that is a separate rank problem.
Finally, the distinction between estimation and testing remains useful: the maintained \(\rho_*<0\) theory does not provide the null distribution for testing \(\rho_*=0\). The estimation results developed in this note should be used within that cointegrated setting.
Estimation and causal identification answer different questions
Estimation asks how to calculate the model’s coefficients and how the resulting errors behave across samples. Cho establishes these statistical properties under the paper’s maintained assumptions. Causal identification asks what makes a coefficient represent the effect of changing an explanatory variable while holding other relevant influences fixed. NARDL estimation does not supply that identification argument by itself. Neither an asymmetric response nor a cointegrating relationship is sufficient to establish causality.
The familiar identification problems in macroeconomics remain. An omitted policy or common shock may move both variables. Measurement error may make the observed regressor differ from the economically relevant variable. Forward-looking households and firms may react to an anticipated policy before its recorded implementation. Reverse causality or simultaneous determination may mean that the outcome also changes the explanatory variable. These mechanisms can make a regressor related to the equation’s unobserved determinants and undermine the exogeneity assumptions used to interpret the estimates.
For example, a coefficient on oil-price increases may reflect both the effect of energy costs and a global-demand shock that raises oil prices and output together. Separating positive and negative oil-price changes does not separate those two mechanisms. A causal interpretation therefore needs a separate, credible argument about the source and timing of the variation, the relevant controls, and the exogeneity restrictions. A larger sample or a two-stage bias correction cannot, by itself, provide that argument.
Main takeaway. Report the adjustment coefficient, dynamic coefficients, and long-run ratios according to the economic question. The OLS versus two-stage choice concerns finite-sample bias, with no universal sample-size cutoff. A causal interpretation additionally requires an identification argument.
Parameter definitions: Appendix A. Same-target comparisons and the long-run ratio: Appendix H.
10. Conclusion
Cho establishes the consistency and limiting distribution of the original OLS estimator in a NARDL model whose normalized regressor matrix is asymptotically singular. The singularity comes from the deterministic trends generated by the positive and negative partial sums. Those trends make several columns proportional under the initial normalization.
The FWL transformations separate the common trend from the stochastic fluctuations and replace the lagged dependent variable with its estimated cointegration residual. Each component can then be normalized at its own scale. The resulting nonsingular matrix allows the transformed estimators’ joint limit to be derived, and the exact coefficient identities recover the distribution of the original OLS estimates.
The reconstruction explains why the original level coefficients generally converge at rate \(\sqrt T\): they contain the error in estimating the adjustment speed. The long-run slopes converge faster. When a population coefficient makes the leading adjustment-error term vanish, the remaining terms determine a different rate.
For applied economists, the results justify the familiar one-step OLS procedure within the cointegrating model studied by Cho. They also clarify how to report the estimates: distinguish the adjustment coefficient, the coefficients in the dynamic equation, and the long-run ratios. In choosing between OLS and two-stage NARDL, the practical issue is finite-sample bias. Direct OLS is convenient in sufficiently large samples; two-stage estimation can be preferable in smaller samples when its bias correction is effective.
Main takeaway. The theoretical contribution is to recover the behavior of the original OLS estimator despite a singular initial matrix limit. Preserving the cumulative stochastic variation before taking that limit makes the argument possible.
References and further reading
Cho, J. S. (2026). “Estimation and Inference of Nonlinear Autoregressive Distributed Lag Models with Time Trend by Ordinary Least Squares.” Econometric Theory, First View. Publisher and DOI · Author’s page · Author’s manuscript. The estimation argument developed here follows Lemma 1, Propositions 1–4, Lemmas 2–4, and Theorem 1.
Shin, Y., Yu, B., and Greenwood-Nimmo, M. (2014). “Modelling Asymmetric Cointegration and Dynamic Multipliers in a Nonlinear ARDL Framework.” In Festschrift in Honor of Peter Schmidt: Econometric Methods and Applications, pp. 281–314. Springer. Publisher and DOI.
Park, J. Y., and Phillips, P. C. B. (1988, 1989). “Statistical Inference in Regression with Integrated Processes,” Parts 1 and 2. Econometric Theory, 4, 468–498, and 5, 95–131. Part 1 · Part 2. These papers develop the integrated-regression projection approach on which Cho builds.
Related EconMacro reading: AR Filtering or Lag Controls? What the Frisch–Waugh–Lovell Theorem Really Says.
Mathematical appendices
Each appendix follows the order of the main argument. Symbols are defined where they first enter a calculation; exact identities are separated from probabilistic limits. The numerical examples illustrate the algebra. The return links take you to the end of the corresponding main-text section, so you can continue with the next section after reading an appendix.
Appendix A. From the levels equation to the error-correction model
Return to the main text: End of Section 2: the error-correction model · End of Section 9: parameter reporting · Contents.
The error-correction representation follows from exact substitutions in the levels equation. This appendix shows those substitutions before introducing the compact vector notation. A star without the sample-size subscript \(T\) denotes a fixed population parameter. Later, \(\vartheta_{T*}\) and \(\tau_{T*}\) denote sample-dependent targets arising from estimated transformations; their definitions retain the subscript \(T\). The shorthand \(q_t=t-1\) denotes the time-trend regressor; the lag order \(q\), which has no time subscript, is a different object.
A.1. A worked example with two lags of the dependent variable
Take one explanatory variable, two lags of \(y_t\), and one lag of each partial sum. The levels equation is
Subtract the observed lagged outcome. Starting from Equation (A.1), subtract one copy of \(y_{t-1}\) from each side. Before collecting any terms, the entire equation becomes
\[ \begin{aligned} y_t-y_{t-1}={}&\alpha_{0*}+\xi_*t +\varphi_{1*}y_{t-1}-y_{t-1}+\varphi_{2*}y_{t-2}\\ &+\theta_{0*}^+x_t^++\theta_{1*}^+x_{t-1}^+ +\theta_{0*}^-x_t^-+\theta_{1*}^-x_{t-1}^-+e_t. \end{aligned} \]The left side is the change in the outcome, \(\Delta y_t=y_t-y_{t-1}\). On the right, the two terms involving \(y_{t-1}\) combine because they multiply the same observed value:
\[ \varphi_{1*}y_{t-1}-y_{t-1} =\varphi_{1*}y_{t-1}-1\times y_{t-1} =(\varphi_{1*}-1)y_{t-1}. \]Consequently,
\[ \begin{aligned} \Delta y_t={}&\alpha_{0*}+\xi_*t +(\varphi_{1*}-1)y_{t-1}+\varphi_{2*}y_{t-2}\\ &+\theta_{0*}^+x_t^++\theta_{1*}^+x_{t-1}^+ +\theta_{0*}^-x_t^-+\theta_{1*}^-x_{t-1}^-+e_t. \end{aligned} \]Express older levels using changes. The change at date \(t-1\) is the value at that date minus the value one period earlier. Rearranging this definition gives
\[ \begin{aligned} \Delta y_{t-1}&=y_{t-1}-y_{t-2},\\ \Delta y_{t-1}+y_{t-2}&=y_{t-1},\\ y_{t-2}&=y_{t-1}-\Delta y_{t-1}. \end{aligned} \]For example, if \(y_{t-1}=12\) and the increase from \(t-2\) to \(t-1\) was \(2\), then \(y_{t-2}=12-2=10\). The identity replaces the older level \(y_{t-2}\) with the more recent level \(y_{t-1}\) and the intervening change.
A partial sum also equals its previous value plus its latest increment. Writing the positive and negative cases separately makes the operation clear:
\[ \begin{aligned} \Delta x_t^+&=x_t^+-x_{t-1}^+ &\Longrightarrow\quad x_t^+&=x_{t-1}^++\Delta x_t^+,\\ \Delta x_t^-&=x_t^–x_{t-1}^- &\Longrightarrow\quad x_t^-&=x_{t-1}^-+\Delta x_t^-. \end{aligned} \]The superscripts identify the positive and negative partial sums. Both obey the same addition rule; a negative increment reduces the negative partial sum. The shorthand \(x_t^\pm=x_{t-1}^\pm+\Delta x_t^\pm\) will henceforth mean these two separate identities. Finally, adding one to \(q_t=t-1\) gives \(t=q_t+1\). These substitutions express the model in lagged levels, changes, and the chosen time regressor.
The lagged dependent-variable terms become
\[ \begin{aligned} (\varphi_{1*}-1)y_{t-1}+\varphi_{2*}y_{t-2} &=(\varphi_{1*}-1)y_{t-1} +\varphi_{2*}(y_{t-1}-\Delta y_{t-1})\\ &=(\varphi_{1*}-1)y_{t-1}+\varphi_{2*}y_{t-1} -\varphi_{2*}\Delta y_{t-1}\\ &=(\varphi_{1*}+\varphi_{2*}-1)y_{t-1} -\varphi_{2*}\Delta y_{t-1}. \end{aligned} \]For each sign, the explanatory-variable terms become
\[ \begin{aligned} \theta_{0*}^{\pm}x_t^\pm+\theta_{1*}^{\pm}x_{t-1}^\pm &=\theta_{0*}^{\pm}(x_{t-1}^\pm+\Delta x_t^\pm) +\theta_{1*}^{\pm}x_{t-1}^\pm\\ &=\theta_{0*}^{\pm}x_{t-1}^\pm +\theta_{0*}^{\pm}\Delta x_t^\pm+\theta_{1*}^{\pm}x_{t-1}^\pm\\ &=(\theta_{0*}^{\pm}+\theta_{1*}^{\pm})x_{t-1}^\pm +\theta_{0*}^{\pm}\Delta x_t^\pm. \end{aligned} \]The time substitution also changes the intercept. Distribute the trend coefficient over \(q_t+1\):
\[ \alpha_{0*}+\xi_*t =\alpha_{0*}+\xi_*(q_t+1) =\alpha_{0*}+\xi_*q_t+\xi_* =(\alpha_{0*}+\xi_*)+\xi_*q_t. \]Collecting the lagged levels, changes, time trend, and constant gives
In Equation (A.2), the disturbance remains \(e_t\). The operation subtracts the observed variable \(y_{t-1}\); it does not subtract the entire model dated \(t-1\). Consequently, it does not produce \(\Delta e_t\).
A.2. From the expanded equation to vector notation
First keep the worked specification \(k=1,p=2,q=1\). Its two partial sums and their coefficients form two-entry columns. Their product is obtained by matching entries and adding:
\[ \begin{aligned} \theta_*^{\prime}\ddot x_{t-1} &=\begin{pmatrix}\theta_*^+&\theta_*^-\end{pmatrix} \begin{pmatrix}x_{t-1}^+\\x_{t-1}^-\end{pmatrix}\\ &=\theta_*^+x_{t-1}^++\theta_*^-x_{t-1}^-. \end{aligned} \]There are three short-run regressors: the lagged change in the outcome, the current positive change, and the current negative change. Put their coefficients in the same order:
\[ z_{2t}=\begin{pmatrix}\Delta y_{t-1}\\\Delta x_t^+\\\Delta x_t^-\end{pmatrix}, \qquad \kappa_*=\begin{pmatrix}\phi_{1*}\\\pi_{0*}^+\\\pi_{0*}^-\end{pmatrix}. \] \[ \begin{aligned} \kappa_*^{\prime}z_{2t} &=\begin{pmatrix}\phi_{1*}&\pi_{0*}^+&\pi_{0*}^-\end{pmatrix} \begin{pmatrix}\Delta y_{t-1}\\\Delta x_t^+\\\Delta x_t^-\end{pmatrix}\\ &=\phi_{1*}\Delta y_{t-1} +\pi_{0*}^+\Delta x_t^+ +\pi_{0*}^-\Delta x_t^-. \end{aligned} \]Both products yield one number at each date. Replacing the corresponding sums in the expanded equation by these products gives
\[ \Delta y_t=\rho_*y_{t-1}+\theta_*^{\prime}\ddot x_{t-1} +\xi_*q_t+\alpha_*+\kappa_*^{\prime}z_{2t}+e_t. \]We can collect all eight regressors and all eight coefficients. The constant regressor is the number one; multiplying it by \(\alpha_*\) supplies the intercept:
\[ z_t=\begin{pmatrix} y_{t-1}\\x_{t-1}^+\\x_{t-1}^-\\q_t\\1\\ \Delta y_{t-1}\\\Delta x_t^+\\\Delta x_t^- \end{pmatrix}, \qquad \boldsymbol\alpha_*=\begin{pmatrix} \rho_*\\\theta_*^+\\\theta_*^-\\\xi_*\\\alpha_*\\ \phi_{1*}\\\pi_{0*}^+\\\pi_{0*}^- \end{pmatrix}. \]Multiplying the coefficient row by the regressor column performs the entire regression calculation for observation \(t\):
\[ \begin{aligned} \boldsymbol\alpha_*^{\prime}z_t &=\begin{pmatrix} \rho_*&\theta_*^+&\theta_*^-&\xi_*&\alpha_*& \phi_{1*}&\pi_{0*}^+&\pi_{0*}^- \end{pmatrix} \begin{pmatrix} y_{t-1}\\x_{t-1}^+\\x_{t-1}^-\\q_t\\1\\ \Delta y_{t-1}\\\Delta x_t^+\\\Delta x_t^- \end{pmatrix}\\[3pt] &=\rho_*y_{t-1}+\theta_*^+x_{t-1}^+ +\theta_*^-x_{t-1}^-+\xi_*q_t+\alpha_*\times1\\ &\quad+\phi_{1*}\Delta y_{t-1} +\pi_{0*}^+\Delta x_t^++\pi_{0*}^-\Delta x_t^-. \end{aligned} \]The compact equation is therefore \(\Delta y_t=z_t^{\prime}\boldsymbol\alpha_*+e_t\). Reversing the two scalar factors in each term shows that \(z_t^{\prime}\boldsymbol\alpha_*\) and \(\boldsymbol\alpha_*^{\prime}z_t\) give the same number. The notation collects the eight contributions already displayed. Bold \(\boldsymbol\alpha_*\) denotes the complete coefficient vector; scalar \(\alpha_*\) denotes its intercept entry.
Allowing more variables and lags. With \(p\) autoregressive lags and explanatory-variable lag order \(q\geq1\), repeated use of the same difference identities gives
\[ \begin{aligned} \rho_*&=\sum_{i=1}^{p}\varphi_{i*}-1, &\phi_{j*}&=-\sum_{i=j+1}^{p}\varphi_{i*}, \quad j=1,\ldots,p-1,\\ \theta_*^\pm&=\sum_{i=0}^{q}\theta_{i*}^\pm, &\pi_{0*}^\pm&=\theta_{0*}^\pm,\\ \pi_{j*}^\pm&=-\sum_{i=j+1}^{q}\theta_{i*}^\pm, &&j=1,\ldots,q-1. \end{aligned} \]For example, a level \(y_{t-3}\) can be written as \(y_{t-1}-\Delta y_{t-1}-\Delta y_{t-2}\). Its coefficient consequently contributes to both lagged-change coefficients. This is why the general definitions add the coefficients on all levels older than the change in question.
With \(k\) original explanatory variables, each signed block has \(k\) entries:
\[ \ddot x_t=\begin{pmatrix}x_t^+\\x_t^-\end{pmatrix}, \qquad \theta_*=\begin{pmatrix}\theta_*^+\\\theta_*^-\end{pmatrix}. \]Thus \(\ddot x_t\) and \(\theta_*\) are \(2k\times1\). The double dot identifies a constructed quantity. The short-run vector contains \(p-1\) lagged outcome changes. For each of the \(2k\) signed explanatory-variable components, it contains \(q\) changes, dated \(t,t-1,\ldots,t-q+1\). Its length is therefore \(s=(p-1)+2kq\). For the worked example, \(s=(2-1)+2\times1\times1=3\), matching the three entries shown above.
The full vectors retain the same block order:
\[ z_t=\begin{pmatrix}y_{t-1}\\\ddot x_{t-1}\\q_t\\1\\z_{2t}\end{pmatrix}, \qquad \boldsymbol\alpha_*=\begin{pmatrix}\rho_*\\\theta_*\\\xi_*\\\alpha_*\\\kappa_*\end{pmatrix}. \]The block lengths are \(1,2k,1,1,s\), giving \(K=2k+s+3\) entries in each vector. The product has dimensions \((1\times K)(K\times1)=1\times1\), as required for the scalar outcome change. In the worked example, \(K=1+2+1+1+3=8\).
A.3. Identify the equilibrium relationship
Define the long-run slopes and trend coefficient by
\[ \beta_*=-\frac{\theta_*}{\rho_*}, \qquad \zeta_*=-\frac{\xi_*}{\rho_*}, \qquad u_{t-1}=y_{t-1}-\beta_*’\ddot x_{t-1}-\zeta_*q_t-\nu_*. \]Dividing the \(2k\times1\) vector \(\theta_*\) by the scalar \(\rho_*\) leaves \(\beta_*\) with dimension \(2k\times1\). The product \(\beta_*’\ddot x_{t-1}\) is scalar. The quantities \(\zeta_*\), \(\nu_*\), and \(u_{t-1}\) are also scalars, so the equation defining the equilibrium error is dimensionally consistent.
The constant \(\nu_*\) centers the cointegration error so that \(E[u_t]=0\). Multiply each term in its defining equation by \(\rho_*\), then use \(-\rho_*\beta_*=\theta_*\) and \(-\rho_*\zeta_*=\xi_*\):
\[ \begin{aligned} \rho_*u_{t-1} &=\rho_*y_{t-1}-\rho_*\beta_*^{\prime}\ddot x_{t-1} -\rho_*\zeta_*q_t-\rho_*\nu_*\\ &=\rho_*y_{t-1}+\theta_*^{\prime}\ddot x_{t-1} +\xi_*q_t-\rho_*\nu_*. \end{aligned} \]Add \(\rho_*\nu_*\) to both sides to recover the three level-and-trend terms appearing in the dynamic equation. Their sum is \(\rho_*u_{t-1}+\rho_*\nu_*\). Combining the latter constant with \(\alpha_*\) produces \(\gamma_*\).
It follows that the dynamic equation can be written as
Equation (A.3) recovers Equation (6) in the main text. This calculation explains the intercept adjustment as well as the signs in the long-run ratios. The coefficient \(\rho_*\) governs adjustment to disequilibrium; the vector \(\beta_*\) describes the equilibrium slopes.
For example, a positive \(u_{t-1}\) means that the observed outcome lies above the level prescribed by the long-run relationship. With \(\rho_*<0\), the contribution \(\rho_*u_{t-1}\) to the next change is negative. The adjustment term therefore moves the outcome back toward equilibrium, holding the other terms fixed. Its size depends on both the distance from equilibrium and the adjustment coefficient.
A.4. A guide to the notation used in the calculations
| Notation | How to read it |
|---|---|
| \(t\), \(T\), \(q_t\), \(q\) | Observation index, sample size, time trend \(t-1\), and explanatory-variable lag order, respectively. |
| A star, hat, tilde, or double dot | A star without the sample-size subscript \(T\) identifies a fixed population parameter. The starred targets \(\vartheta_{T*}\) and \(\tau_{T*}\) depend on the estimated transformation, as their \(T\) subscript records. Hats identify the original one-step OLS estimates. Tildes and double dots distinguish auxiliary estimates and transformed quantities. A double dot is never a time derivative here. |
| \(\theta_*\) and \(\beta_*\) | The dynamic equation’s level coefficients and the equilibrium slopes, linked by \(\theta_*=-\rho_*\beta_*\). |
| \(\rho_*\) and \(\varrho_*\) | The adjustment coefficient and the different lag-polynomial quantity defined in Appendix B.5. |
| \(\nu\) and \(\upsilon\) | Greek nu denotes the long-run intercept. Greek upsilon denotes a vector of long-run regression coefficients. |
| \(\vartheta\) and \(\varsigma\) | The transformed trend coefficients in the long-run and dynamic regressions, respectively. |
| \(D\), \(\ddot D\), and \(\mathcal D\) | The first two are normalization matrices. Calligraphic \(\mathcal D\) is a random vector describing a limiting distribution. |
| \(\xrightarrow{p}\) and \(\Rightarrow\) | Convergence in probability and convergence in distribution. In the fixed-target probability limits used here, the chance of a discrepancy larger than any chosen positive tolerance tends to zero. Convergence in distribution describes the limiting probabilities of a random quantity, such as an estimation error multiplied by \(T\) or \(\sqrt T\) to keep its shrinking fluctuations visible. |
| \(X_T=O_p(a_T)\) | The ratio \(X_T/a_T\) stays bounded in probability. The scale \(a_T\) bounds its stochastic magnitude; the notation does not assert an exact equality or a nonzero limiting fluctuation. |
| \(X_T=o_p(a_T)\) | The ratio \(X_T/a_T\) converges to zero in probability. Thus \(o_p(1)\) denotes a term that vanishes. |
Table notes. Each row distinguishes symbols or notation serving different purposes. For example, \(\ddot\beta_T\) is an auxiliary estimate of the population vector \(\beta_*\): the double dot identifies the estimation representation, and the subscript \(T\) records its dependence on the observed sample. Transposing a vector changes its orientation, while these notation labels leave its dimensions unchanged.
Reading a stochastic order. In \(X_T=O_p(a_T)\), first divide the random quantity by its proposed scale: \(X_T/a_T\). “Bounded in probability” means that, for any chosen \(\varepsilon>0\), a sufficiently wide fixed interval contains this ratio with probability at least \(1-\varepsilon\) for all sufficiently large samples. The ratio need not stay inside one fixed interval in every possible sample.
For a simple illustration, let \(U\) have a fixed standard-normal distribution and put \(X_T=U/\sqrt T\). Then \(X_T/(T^{-1/2})=U\): the scaled distribution stays the same as \(T\) grows. Thus \(X_T=O_p(T^{-1/2})\). If instead \(Y_T=U/T\), then \(Y_T/(T^{-1/2})=U/\sqrt T\xrightarrow{p}0\), so \(Y_T=o_p(T^{-1/2})\). The second quantity becomes negligible even relative to the first scale. These examples illustrate the notation; the paper derives its own coefficient distributions. The main discussion in Section 3 introduces the growth scales used below.
Dimensions and multiplication. A dimension is written as rows \(\times\) columns. Two matrices can be multiplied when the number of columns of the first equals the number of rows of the second. The result retains the first matrix’s number of rows and the second matrix’s number of columns. For example, \((K\times1)(1\times K)\) produces a \(K\times K\) matrix, while \((1\times K)(K\times1)\) produces a scalar.
| Object | Dimensions | What the rows and columns represent |
|---|---|---|
| \(\ddot x_{t-1}\), \(\theta_*\), \(\beta_*\), \(\mu_*\) | \(2k\times1\) | One entry for each positive or negative component. |
| \(z_{2t}\), \(\kappa_*\) | \(s\times1\) | One entry for each short-run regressor or its coefficient. |
| \(z_t\), \(\boldsymbol\alpha_*\) | \(K\times1\) | One entry for every regressor or coefficient in the full equation. |
| \(Z\), with row \(t\) equal to \(z_t’\) | \(T\times K\) | \(T\) observations in rows and \(K\) regressors in columns. |
| \(Z’Z=\sum_tz_tz_t’\) | \(K\times K\) | One row and one column for every regressor; each entry sums a pair of regressor products. |
Table notes. The first three rows concern the variables or coefficients for one date. The fourth row stacks the observation rows over all \(T\) dates. The final row multiplies \(Z^{\prime}\), which has \(K\) rows and \(T\) columns, by \(Z\), which has \(T\) rows and \(K\) columns. Each resulting entry adds \(T\) products, one per observation. The resulting \(K\times K\) matrix therefore contains a sum of products for every pair of regressors. The following smaller example shows exactly what that multiplication does.
Take three observations and only two regressors, a constant and a trend taking values \(0,1,2\). This is an illustration of matrix multiplication. Its design matrix and transpose are
\[ Z=\begin{pmatrix}1&0\\1&1\\1&2\end{pmatrix}, \qquad Z^{\prime}=\begin{pmatrix}1&1&1\\0&1&2\end{pmatrix}. \]The transpose turns each regressor column into a row. To obtain one entry of \(Z^{\prime}Z\), multiply the three corresponding row and column entries, then add:
\[ \begin{aligned} Z^{\prime}Z &=\begin{pmatrix}1&1&1\\0&1&2\end{pmatrix} \begin{pmatrix}1&0\\1&1\\1&2\end{pmatrix}\\ &=\begin{pmatrix} 1(1)+1(1)+1(1)&1(0)+1(1)+1(2)\\ 0(1)+1(1)+2(1)&0(0)+1(1)+2(2) \end{pmatrix}\\ &=\begin{pmatrix}3&3\\3&5\end{pmatrix}. \end{aligned} \]The top-left entry adds the three squared ones. The bottom-right entry adds the three squared trend values, \(0^2+1^2+2^2=5\). The off-diagonal entries both add the constant–trend products, \(0+1+2=3\). Every entry is one scalar, and the four entries together form a \(2\times2\) matrix. Its two rows and columns identify the two regressors; the three observations have been added within each entry.
Returning to the NARDL example with \(T=100\) and \(K=8\), the same operation multiplies an \(8\times100\) matrix by a \(100\times8\) matrix and produces an \(8\times8\) cross-product matrix. Each of its 64 entries adds 100 regressor products. Adding observations lengthens these sums, while the matrix remains \(8\times8\) as long as the specification retains eight regressors.
Main takeaway. The error-correction representation follows from exact substitutions in the levels equation. Each inner product is scalar, and the complete regressor vector has \(K=2k+s+3\) entries.
Return to the main text: End of Section 2: the error-correction model · End of Section 9: parameter reporting · Contents
Appendix B. Why partial sums drift, and where the powers of T come from
Return to the main text: End of Section 3: drift and stochastic fluctuations · End of Section 5: column normalizations · End of Section 4: normalization and vanishing terms · Contents.
B.1. Separating increases and decreases changes their means
For a scalar change \(v=\Delta x_t\), define \(v^+=\max(v,0)\) and \(v^-=\min(v,0)\). The negative part retains its sign. Two useful identities are
\[ v^+=\frac{v+|v|}{2}, \qquad v^-=\frac{v-|v|}{2}. \]These identities can be checked by considering the sign of \(v\). If \(v\geq0\), then \(|v|=v\), so \((v+|v|)/2=v\) and \((v-|v|)/2=0\). If \(v<0\), then \(|v|=-v\), so the first expression is zero and the second is \((v+v)/2=v\). This reproduces the positive and negative parts in both cases. Taking expectations gives
\[ \mu_*^+=\frac{E[v]+E|v|}{2}, \qquad \mu_*^-=\frac{E[v]-E|v|}{2}. \]If \(E[v]=0\) and changes are nondegenerate, then \(\mu_*^+=E|v|/2>0\) and \(\mu_*^-=-E|v|/2<0\). Zero mean for the original change therefore coexists with positive and negative means for its two components.
For a simple calculation, let a change take the values \(-2,-1,+1,+2\), each with probability one quarter. The original mean is \((-2-1+1+2)/4=0\). The positive part takes the values \(0,0,1,2\), giving mean \((0+0+1+2)/4=3/4\). The negative part takes the values \(-2,-1,0,0\), giving mean \((-2-1+0+0)/4=-3/4\). The component means cancel when added, although each signed partial sum has its own nonzero drift.
The partial sum up to date \(t\) contains \(t\) increments. If their common mean is \(\mu_*^+\), linearity of expectation gives
\[ \begin{aligned} E(x_t^+) &=E\!\left(\sum_{j=1}^{t}\Delta x_j^+\right) =\sum_{j=1}^{t}E(\Delta x_j^+)\\ &=\underbrace{\mu_*^++\cdots+\mu_*^+}_{t\text{ copies}} =t\mu_*^+. \end{aligned} \]For the negative series the same addition gives \(E(x_t^-)=t\mu_*^-\). In the numerical example, these means are \(3t/4\) and \(-3t/4\). No independence assumption is needed for adding expectations. The factor \(t\) counts the increments accumulated by date \(t\), while \(T\) denotes the full sample size.
Define centered increments \(s_j^\pm=\Delta x_j^\pm-\mu_*^\pm\). To see the deterministic and random parts separately, substitute \(\Delta x_j^\pm=\mu_*^\pm+s_j^\pm\) for every increment:
Equation (B.1) expands the decomposition in Equation (7): the deterministic part is the sum of the means; the random part is the sum of departures from those means.
In vector form, \(\ddot x_t=\mu_*t+m_t\). Each of \(\ddot x_t\), \(\mu_*\), and \(m_t\) is \(2k\times1\); multiplying the drift vector \(\mu_*\) by the scalar time index \(t\) preserves that dimension. The deterministic component grows linearly with time; the centered cumulative component fluctuates around it. Appendix B.2 isolates one scalar component to explain its variance.
B.2. Why a centered cumulative process has a square-root scale
To isolate the arithmetic, consider independent increments \(\eta_j\) with zero mean and variance \(\sigma^2\), and let \(m_t=\sum_{j=1}^{t}\eta_j\). Its mean is zero. Its variance is
The covariance terms in Equation (B.2) vanish under independence. Standard deviation is the square root of variance, so
\[ \operatorname{SD}(m_t)=\sigma\sqrt t. \]Variances add across independent increments. This produces a standard deviation proportional to \(\sqrt t\), even though there are \(t\) increments. For \(\sigma=1\), the standard deviation is \(10\) after \(100\) increments and \(20\) after \(400\).
With nonzero drift \(\mu\), the relative stochastic scale is
\[ \frac{\sigma\sqrt t}{|\mu|t} =\frac{\sigma}{|\mu|\sqrt t}\longrightarrow0. \]The cumulative fluctuations continue to grow in absolute size, while becoming smaller relative to the deterministic drift. Under suitable short-memory dependence, an analogous calculation uses the long-run variance of the increments. The paper’s assumptions provide the functional limit theory needed for its processes.
B.3. From an observation’s size to a column’s normalization
With \(T\) observations, the time index \(t\) runs from \(1\) to \(T\). The trend regressor \(q_t=t-1\) therefore takes the values
\[ q_1=0,\quad q_2=1,\quad q_3=2,\quad\ldots,\quad q_T=T-1. \]There are \(T\) entries in this column, including the initial zero. We need two sums. Write \(S_1\) for the sum of the entries and \(S_2\) for the sum of their squares:
\[ \begin{aligned} S_1&=\sum_{t=1}^{T}q_t=0+1+2+\cdots+(T-1),\\ S_2&=\sum_{t=1}^{T}q_t^2=0^2+1^2+2^2+\cdots+(T-1)^2. \end{aligned} \]These sums enter the OLS regressor cross-product matrix. Multiplying the constant column by the trend column gives \(S_1\), because every entry of the constant column equals one. Multiplying the trend column by itself gives \(S_2\). Thus, \(S_2\) is the diagonal entry associated with the trend, while \(S_1\) appears in the two off-diagonal positions linking the trend and the constant.
Calculating the sum of the trend values. Write \(S_1\) twice, with the second line in reverse order:
\[ \begin{aligned} S_1&=0+1+\cdots+(T-2)+(T-1),\\ S_1&=(T-1)+(T-2)+\cdots+1+0. \end{aligned} \]Add the two lines term by term. The first pair is \(0+(T-1)=T-1\), the second is \(1+(T-2)=T-1\), and every remaining pair also equals \(T-1\). There are \(T\) pairs, so
Calculating the sum of squared trend values. Squaring each entry before adding produces a different sum. To derive its formula, start with the expansion of a cube:
\[ (j+1)^3=j^3+3j^2+3j+1, \qquad (j+1)^3-j^3=3j^2+3j+1. \]Here \(j\) runs through the trend values \(0,1,\ldots,T-1\). Add this identity over those \(T\) values. The final two brackets are worth writing explicitly:
\[ \begin{aligned} \sum_{j=0}^{T-1}\bigl[(j+1)^3-j^3\bigr] &=(1^3-0^3)+(2^3-1^3)+\cdots\\ &\quad+\bigl[(T-1)^3-(T-2)^3\bigr] +\bigl[T^3-(T-1)^3\bigr]. \end{aligned} \]The \(+1^3\) in the first bracket cancels the \(-1^3\) in the second. At the other end, the \(+(T-1)^3\) in the penultimate bracket cancels the \(-(T-1)^3\) in the last bracket. Every intermediate cube occurs once positively and once negatively. This is a telescoping sum: those adjacent cancellations leave only the endpoints.
With \(T=4\), every term can be shown:
\[ \begin{aligned} &(1^3-0^3)+(2^3-1^3)+(3^3-2^3)+(4^3-3^3)\\ &\quad=-0^3+(1^3-1^3)+(2^3-2^3)+(3^3-3^3)+4^3\\ &\quad=4^3=64. \end{aligned} \]For a general \(T\), only \(T^3-0^3=T^3\) remains. There is no \(-T^3\) term because the sum stops at \(j=T-1\); that negative term would belong to the next, excluded bracket \((T+1)^3-T^3\). On the right side of the cube identity, adding \(3j^2\) gives \(3S_2\), adding \(3j\) gives \(3S_1\), and adding \(T\) copies of one gives \(T\). Hence
\[ T^3=3S_2+3S_1+T. \]Subtract \(3S_1+T\), divide by three, and substitute Equation (B.3) for \(S_1\):
\[ \begin{aligned} S_2 &=\frac{T^3-3S_1-T}{3}\\ &=\frac{T^3-\frac32T(T-1)-T}{3}\\ &=\frac{2T^3-3T(T-1)-2T}{6}\\ &=\frac{2T^3-3T^2+T}{6}\\ &=\frac{T(2T^2-3T+1)}6. \end{aligned} \]The quadratic in the last line factors as \(2T^2-3T+1=(T-1)(2T-1)\). This gives the second formula:
To check both Equation (B.3) and Equation (B.4), take \(T=4\). The trend column is \((0,1,2,3)’\), and direct addition gives
\[ S_1=0+1+2+3=6, \qquad S_2=0^2+1^2+2^2+3^2=14. \]The formulas give exactly the same numbers:
\[ \frac{4(4-1)}2=6, \qquad \frac{(4-1)\times4\times(2\times4-1)}6=14. \]Why the sum of squares grows like \(T^3\). Expanding the exact expression separates the three powers of sample size:
\[ S_2=\frac{T^3}{3}-\frac{T^2}{2}+\frac{T}{6}. \]To compare their importance as \(T\) grows, divide every term by \(T^3\):
\[ \frac{S_2}{T^3} =\frac13-\frac1{2T}+\frac1{6T^2} \longrightarrow\frac13. \]The second and third terms tend to zero. Thus, saying that \(S_2\) is approximately \(T^3/3\) means that the ratio \(S_2/(T^3/3)\) approaches one. The growing power is \(T^3\); the factor \(1/3\) is the constant that remains after normalization.
Why the column is divided by \(T^{3/2}\). The sum of squares is built by multiplying each column entry by itself. If every entry \(q_t\) is divided by a number \(\lambda_T\), its square is divided by \(\lambda_T^2\):
\[ \sum_{t=1}^{T}\left(\frac{q_t}{\lambda_T}\right)^2 =\frac{S_2}{\lambda_T^2}. \]Since \(S_2\) grows like \(T^3\), choose \(\lambda_T^2=T^3\). Taking the positive square root gives \(\lambda_T=\sqrt{T^3}=T^{3/2}=T\sqrt T\). Consequently,
\[ \sum_{t=1}^{T}\left(\frac{q_t}{T^{3/2}}\right)^2 =\frac{1}{T^3}\sum_{t=1}^{T}q_t^2 \longrightarrow\frac13. \]This is what the column normalization accomplishes: the trend’s diagonal entry in the normalized OLS matrix approaches a finite, positive number. The factor \(T^{3/2}\) scales the column, and its square \(T^3\) scales the corresponding sum of squares.
Applying the same reasoning to the other columns. In the example from Appendix B.2, the centered cumulative process \(m_{t-1}\) has mean zero and variance \(\sigma^2(t-1)\). Its expected square therefore equals that variance. Adding across observations gives
\[ E\left[\sum_{t=1}^{T}m_{t-1}^{\,2}\right] =\sigma^2\sum_{t=1}^{T}(t-1) =\frac{\sigma^2T(T-1)}2. \]The formula for \(S_1\) now explains the last equality: the expected squared cumulative fluctuations add up to \(\sigma^2\) times \(0+1+\cdots+(T-1)\). This expected sum grows like \(T^2\). Under the functional limit assumptions discussed above, the random sum of squares has that scale, giving column normalization \(\sqrt{T^2}=T\).
For the constant column, each entry and each squared entry equals one. There are \(T\) observations, so its sum of squares is exactly \(T\). Dividing the column by \(\sqrt T\) therefore gives
\[ \sum_{t=1}^{T}\left(\frac{1}{\sqrt T}\right)^2 =\sum_{t=1}^{T}\frac1T =1. \]How to read the table. This table supplies the calculations behind the column-normalization table in Section 5. Each row follows one type of regressor from the magnitude of an individual observation to the sum of squared observations, then to the divisor applied to its entire column. The entries give growth orders as the sample size \(T\) increases. Constants such as \(\sigma\) and \(1/3\) are omitted from these orders; their values remain in the formulas above.
The observation column compares values at the same relative position in expanding samples. For example, halfway through a sample of \(100\) observations, \(t=50\); halfway through a sample of \(400\), \(t=200\). In both cases \(t=T/2\). This explains the earlier phrase “sample times proportional to \(T\)”: the time index grows with the sample size. An early observation such as \(q_2=1\) stays fixed when later observations are added.
| Regressor | Magnitude of one observation | Order of the sum of squares | Divide the column by |
|---|---|---|---|
| Linear trend | \(T\) | \(T^3\) | \(T^{3/2}\) |
| Centered cumulative fluctuations | \(\sqrt T\) | \(T^2\) | \(T\) |
| Stationary regressor or constant | \(1\) | \(T\) | \(\sqrt T\) |
Where the additional factor \(T\) enters. The sum of squared observations equals their average multiplied by their number:
\[ \sum_{t=1}^{T}w_t^2 =T\left(\frac1T\sum_{t=1}^{T}w_t^2\right). \]For the trend, the average squared observation has order \(T^2\); adding all \(T\) observations gives order \(T\times T^2=T^3\). For centered cumulative fluctuations, the corresponding average has stochastic order \(T\), giving a sum of order \(T^2\). For a stationary regressor with a finite second moment, the average stays on a fixed scale, giving a sum of order \(T\). The calculations above establish these orders; counting terms alone would not establish a rate for a signed sum in which terms can cancel.
First row: the linear trend. At a time proportional to \(T\), a trend value has magnitude proportional to \(T\). Squaring gives magnitude \(T^2\); summing across the sample gives order \(T^3\). The exact calculation above establishes the factor \(1/3\). The sequence of powers is
\[ T\quad\longrightarrow\quad T^2\quad\longrightarrow\quad T\times T^2=T^3\quad\longrightarrow\quad\sqrt{T^3}=T^{3/2}. \]These four steps are the magnitude of one observation, its square, the sum of squares, and the divisor for the column. The last step takes a square root because dividing each observation by a divisor divides its square by the square of that divisor.
Second row: centered cumulative fluctuations. These are the accumulated random changes left after removing their mean drift. In the independent-increment example, their standard deviation grows as \(\sqrt T\) at times proportional to \(T\). This describes a typical stochastic magnitude: an individual realization can lie above or below that scale. Squaring gives magnitude \(T\), and the sum of squares has order \(T^2\). Thus,
\[ \sqrt T\quad\longrightarrow\quad T\quad\longrightarrow\quad T\times T=T^2\quad\longrightarrow\quad\sqrt{T^2}=T. \]Dividing the cumulative-process column by \(T\) keeps its sum of squares on a finite scale. Under the functional limit theory, this normalized sum generally converges in distribution to a random quantity.
Third row: a stationary regressor or the constant. The entry \(1\) means that the typical magnitude of an observation remains on a fixed scale as the sample grows. A stationary regressor can take many values; the entry describes its growth order. With a finite, positive second moment and the relevant law of large numbers, its average squared value approaches a finite, positive constant. Its sum of squares therefore grows in proportion to \(T\). For the constant column, every observation is exactly one. The powers are
\[ 1\quad\longrightarrow\quad1\quad\longrightarrow\quad T\times1=T\quad\longrightarrow\quad\sqrt T. \]For example, when \(T=100\), the divisors in the final column are \(100^{3/2}=1000\) for the trend, \(100\) for the centered cumulative process, and \(\sqrt{100}=10\) for a stationary regressor or the constant. Each divisor applies to every observation in its column. These normalizations are used to study the OLS matrix as the sample grows; the convergence rates of the original coefficient estimates also depend on the error terms and the coefficient transformations.
A cross-product uses the product of the two column scales. Thus a trend–constant entry is divided by \(T^{3/2}\sqrt T=T^2\); a trend–trend entry is divided by \(T^3\); and a constant–constant entry is divided by \(T\).
For example, multiply the normalized constant by the normalized trend at each observation and then add. The formula for \(S_1\) gives
\[ \begin{aligned} \sum_{t=1}^{T}\frac{1}{\sqrt T}\frac{q_t}{T^{3/2}} &=\frac{1}{T^2}\sum_{t=1}^{T}q_t\\ &=\frac{T(T-1)}{2T^2} =\frac12-\frac1{2T} \longrightarrow\frac12. \end{aligned} \]The normalized constant’s diagonal entry is therefore \(1\), the trend’s diagonal entry tends to \(1/3\), and their cross-product tends to \(1/2\). These are the entries calculated again within the larger OLS matrix in Appendix C.
B.4. Signed error products require their own calculation
A sum of regressor–error products contains positive and negative contributions. Its size cannot be obtained by simply adding the absolute sizes of its terms. For independent mean-zero errors with variance \(\sigma_e^2\), a deterministic trend gives
\[ \operatorname{Var}\left(\sum_{t=1}^{T}q_te_t\right) =\sigma_e^2\sum_{t=1}^{T}q_t^2, \]so this signed sum has standard-deviation scale \(T^{3/2}\), whereas the trend’s sum of squares has scale \(T^3\). Similarly, with independent errors conditional on a cumulative regressor path, the conditional variance of \(\sum_t m_{t-1}e_t\) is \(\sigma_e^2\sum_t m_{t-1}^2\), giving scale \(T\).
These examples explain the arithmetic behind the normalized scores. The full multivariate result additionally accounts for cross-products, dependence, and generated regressors. A column normalization alone does not establish a coefficient’s convergence rate around a fixed population parameter.
B.5. The dependent variable also has a drift
Cho defines \(\delta_*=E[\Delta y_t]\), and then centers the change as \(d_t=\Delta y_t-\delta_*\). Summing gives
\[ y_t=y_0+\delta_*t+\sum_{j=1}^{t}d_j. \]The equilibrium relationship also implies \(\delta_*=\beta_*’\mu_*+\zeta_*\). Taking expectations in the dynamic equation provides another expression:
\[ \delta_*= \frac{\displaystyle \gamma_*+ \sum_{j=0}^{q-1} \left(\pi_{j*}^{+\prime}\mu_*^+ +\pi_{j*}^{-\prime}\mu_*^-\right)} {\displaystyle 1-\sum_{j=1}^{p-1}\phi_{j*}}. \]The denominator is the paper’s \(\varrho_*=1-\sum_j\phi_{j*}\). It is distinct from the adjustment coefficient \(\rho_*\). This identity follows by collecting the lagged mean changes, each equal to \(\delta_*\), on the left side of the expectation equation.
In this expectation calculation, \(\pi_{j*}^+\) and \(\mu_*^+\) are \(k\times1\) vectors, as are their negative counterparts. Each product \(\pi_{j*}^{+\prime}\mu_*^+\) or \(\pi_{j*}^{-\prime}\mu_*^-\) multiplies a \(1\times k\) row by a \(k\times1\) column and gives a scalar. The numerator, denominator, and resulting mean change \(\delta_*\) are therefore all scalars.
More explicitly, the expectation equation contains \(\delta_*\) on the left and \(\sum_j\phi_{j*}\delta_*\) on the right. Subtract the latter from both sides to obtain \((1-\sum_j\phi_{j*})\delta_*\). The cointegration-error contribution disappears because its mean is zero. The signed changes contribute their respective means, which explains the remaining terms in the numerator.
Main takeaway. Trend, cumulative-fluctuation, and stationary columns have sums of squares of orders \(T^3\), \(T^2\), and \(T\), respectively. Dividing each column by the square root of its sum-of-squares order explains the normalizers \(T^{3/2}\), \(T\), and \(\sqrt T\).
Return to the main text: End of Section 3: drift and stochastic fluctuations · End of Section 5: column normalizations · End of Section 4: normalization and vanishing terms · Contents
Appendix C. Locate the singularity inside the OLS matrix
Return to the main text: End of Section 4: the singular limiting matrix · End of Section 1: motivation · Contents.
C.1. Normalize each row and column explicitly
Collect the drifting regressors in \(g_t=(y_{t-1},\ddot x_{t-1}’,q_t)’\). This is the vector called \(v_t\) in Section 4. It contains \(d=1+2k+1=2k+2\) entries and has dimension \(d\times1\). The drift vector \(a=(\delta_*,\mu_*’,1)’\) also has dimension \(d\times1\), so its product with the scalar \(q_t\) has the right dimension to describe the leading component of \(g_t\).
Write \(z_t=(g_t’,1,z_{2t}’)’\). Its three blocks have \(d\), \(1\), and \(s\) rows, totaling \(K=d+1+s=2k+s+3\). Define
To see what the left and right multiplications in Equation (C.1) do, first keep just the trend and constant columns, in that order. Write their \(T\times2\) design matrix and its transpose:
\[ Z_{qc}=\begin{pmatrix} 0&1\\1&1\\\vdots&\vdots\\T-1&1 \end{pmatrix}, \qquad Z_{qc}’=\begin{pmatrix} 0&1&\cdots&T-1\\ 1&1&\cdots&1 \end{pmatrix}. \]Multiplying the \(2\times T\) transpose by the \(T\times2\) design matrix produces a \(2\times2\) cross-product matrix. Each entry adds the products from one row and one column:
\[ G_{qc}=Z_{qc}’Z_{qc} =\begin{pmatrix} \sum_tq_t^2&\sum_tq_t\cdot1\\ \sum_t1\cdot q_t&\sum_t1^2 \end{pmatrix} =\begin{pmatrix}S_2&S_1\\S_1&T\end{pmatrix}. \]Here \(S_1=\sum_{t=1}^{T}q_t\) and \(S_2=\sum_{t=1}^{T}q_t^2\), as in Appendix B.3. The first row and column concern the trend; the second concern the constant. The two off-diagonal entries agree because \(q_t\cdot1=1\cdot q_t\). The normalizer and its inverse are
\[ D_{qc}=\begin{pmatrix}T^{3/2}&0\\0&\sqrt T\end{pmatrix}, \qquad D_{qc}^{-1}=\begin{pmatrix}T^{-3/2}&0\\0&T^{-1/2}\end{pmatrix}. \]First multiply on the left. The four row-by-column calculations are
\[ \begin{aligned} (D_{qc}^{-1}G_{qc})_{11}&=T^{-3/2}S_2+0S_1=S_2/T^{3/2},\\ (D_{qc}^{-1}G_{qc})_{12}&=T^{-3/2}S_1+0T=S_1/T^{3/2},\\ (D_{qc}^{-1}G_{qc})_{21}&=0S_2+T^{-1/2}S_1=S_1/\sqrt T,\\ (D_{qc}^{-1}G_{qc})_{22}&=0S_1+T^{-1/2}T=T/\sqrt T. \end{aligned} \]Collecting them gives the intermediate matrix
\[ D_{qc}^{-1}G_{qc} =\begin{pmatrix} S_2/T^{3/2}&S_1/T^{3/2}\\ S_1/\sqrt T&T/\sqrt T \end{pmatrix}. \]Thus left multiplication divides the first row by \(T^{3/2}\) and the second by \(\sqrt T\). This intermediate matrix need not be symmetric because the two rows have received different divisors.
Now multiply on the right. Denote the final matrix by \(H_T=D_{qc}^{-1}G_{qc}D_{qc}^{-1}\). The four calculations are
\[ \begin{aligned} (H_T)_{11}&=\frac{S_2}{T^{3/2}}T^{-3/2}+\frac{S_1}{T^{3/2}}0 =\frac{S_2}{T^3},\\ (H_T)_{12}&=\frac{S_2}{T^{3/2}}0+\frac{S_1}{T^{3/2}}T^{-1/2} =\frac{S_1}{T^2},\\ (H_T)_{21}&=\frac{S_1}{\sqrt T}T^{-3/2}+\frac{T}{\sqrt T}0 =\frac{S_1}{T^2},\\ (H_T)_{22}&=\frac{S_1}{\sqrt T}0+\frac{T}{\sqrt T}T^{-1/2} =\frac{T}{T}=1. \end{aligned} \]Right multiplication divides the first column by \(T^{3/2}\) and the second by \(\sqrt T\). Every entry has now received the divisor for its row and the divisor for its column. The exponents add when powers are multiplied:
\[ T^{3/2}T^{3/2}=T^{3},\qquad T^{3/2}T^{1/2}=T^{2},\qquad T^{1/2}T^{1/2}=T. \]The final matrix is symmetric, and the exact sums from Appendix B.3 give its limit:
\[ \begin{aligned} H_T &=\begin{pmatrix} S_2/T^3&S_1/T^2\\S_1/T^2&1 \end{pmatrix}\\ &=\begin{pmatrix} \frac13-\frac1{2T}+\frac1{6T^2}&\frac12-\frac1{2T}\\ \frac12-\frac1{2T}&1 \end{pmatrix} \longrightarrow \begin{pmatrix}1/3&1/2\\1/2&1\end{pmatrix}. \end{aligned} \]Why the normalizer appears twice. We normalize each data column once, forming \(Z_{qc}D_{qc}^{-1}\). The cross-product matrix then multiplies the transpose of this normalized design matrix by the normalized design matrix itself:
\[ (Z_{qc}D_{qc}^{-1})'(Z_{qc}D_{qc}^{-1}) =D_{qc}^{-1}Z_{qc}’Z_{qc}D_{qc}^{-1}=H_T. \]Each cross-product contains two normalized regressors, which accounts for the two divisors. The inverse normalizer equals its transpose because it is diagonal. As a numerical check, \(T=4\) gives \(S_2=14\), \(S_1=6\), and column divisors \(8\) and \(2\). Therefore
\[ H_4=\begin{pmatrix}14/(8\times8)&6/(8\times2)\\6/(2\times8)&4/(2\times2)\end{pmatrix} =\begin{pmatrix}14/64&6/16\\6/16&1\end{pmatrix}. \]The trend-and-constant limiting matrix is nonsingular: its determinant is \((1/3)\times1-(1/2)^2=1/12>0\). The full expression \(A_T\) applies the same normalization to every regressor pair. Its singularity arises from the additional drifting columns, whose limiting columns become proportional, as Section 4 showed.
Here \(I_{2k+2}=I_d\) is a \(d\times d\) identity matrix, and \(I_s\) is an \(s\times s\) identity matrix. An identity matrix has ones on its diagonal and zeros elsewhere. The notation \(\operatorname{diag}\) places the displayed square blocks along one diagonal, with zero blocks between them. Thus \(D\), \(D^{-1}\), \(\sum_tz_tz_t’\), and \(A_T\) are all \(K\times K\). The normalization changes their entries while preserving their dimensions.
Let \(h=E[z_{2t}]\), an \(s\times1\) vector of short-run regressor means, and \(M_{22}=E[z_{2t}z_{2t}’]\), an \(s\times s\) matrix of second moments. In this ordering, Lemma 1 gives the full \(K\times K\) limit
In Equation (C.2), the three row and column groups are the drifting regressors, the constant, and the stationary short-run regressors. The stationary variables can have nonzero means; this is why \(h\) appears.
The matrix statement \(A_T\xrightarrow{p}M\) applies entry by entry. The entries of \(A_T\) vary with the sample; the corresponding entries of \(M\) are fixed numbers determined by population drift coefficients and stationary moments. With the number of regressors fixed, the probability that any entry differs from its target by more than a chosen tolerance tends to zero:
\[ \Pr\!\left(\max_{i,j}|(A_T)_{ij}-M_{ij}|>\varepsilon\right) \longrightarrow0 \qquad\text{for every }\varepsilon>0. \]The maximum simply checks the largest entrywise discrepancy. The limiting matrix can be singular even though all these approximations become accurate. Its drifting columns approach proportional columns, so the limiting matrix loses the independent variation needed for inversion. The inverse of a finite-sample matrix therefore cannot be obtained here by inverting \(M\).
The displayed array has three block rows and three block columns. Its full dimensions are \(K\times K\), because most displayed entries are themselves matrices. Their dimensions are
| Row group | Drifting columns: \(d\) | Constant column: \(1\) | Short-run columns: \(s\) |
|---|---|---|---|
| Drifting rows: \(d\) | \(aa’/3:\ d\times d\) | \(a/2:\ d\times1\) | \(ah’/2:\ d\times s\) |
| Constant row: \(1\) | \(a’/2:\ 1\times d\) | \(1:\ 1\times1\) | \(h’:\ 1\times s\) |
| Short-run rows: \(s\) | \(ha’/2:\ s\times d\) | \(h:\ s\times1\) | \(M_{22}:\ s\times s\) |
Table notes. Choose a row group and a column group to identify the pair of regressor types. The upper-right block joins drifting regressors to short-run regressors, while the lower-left block reverses their order and is its transpose. Here \(d=2k+2\), \(s\) counts short-run regressors, and \(K=d+1+s\). Every block in a block row must have the same number of rows; every block in a block column must have the same number of columns. Thus the three-by-three arrangement contains \(K\) actual rows and \(K\) actual columns.
Work through the upper-right block with the example \(k=1,p=2,q=1\). The four drifting regressors are \(y_{t-1},x_{t-1}^+,x_{t-1}^-,q_t\); the three short-run regressors are \(\Delta y_{t-1},\Delta x_t^+,\Delta x_t^-\). Write their means as \(h_1,h_2,h_3\), so \(h=E[z_{2t}]=(h_1,h_2,h_3)’\). Multiplying the four-entry column by the three-entry row gives
\[ \begin{aligned} \frac12ah’ &=\frac12 \begin{pmatrix}\delta_*\\\mu_*^+\\\mu_*^-\\1\end{pmatrix} \begin{pmatrix}h_1&h_2&h_3\end{pmatrix}\\ &=\frac12 \begin{pmatrix} \delta_*h_1&\delta_*h_2&\delta_*h_3\\ \mu_*^+h_1&\mu_*^+h_2&\mu_*^+h_3\\ \mu_*^-h_1&\mu_*^-h_2&\mu_*^-h_3\\ h_1&h_2&h_3 \end{pmatrix}. \end{aligned} \]Each entry is one scalar product of two numbers. For example, row 2, column 3 of this block joins the lagged positive partial sum to the current negative change. Its limit is \(\mu_*^+h_3/2\): the drift of the first regressor times the mean of the second, divided by two. The factor \(1/2\) comes from the normalized sum of time indices calculated in C.2. In the full eight-by-eight matrix, this cell is at row 2, column 8.
This is a column multiplied by a row: it creates all twelve pairings and therefore produces a \(4\times3\) matrix. In Section 2, a coefficient row multiplied by a regressor column instead added matching products and produced one scalar. The order and orientation of the vectors determine which operation is being performed.
C.2. Calculate representative entries
The explicit trend’s diagonal entry is its normalized sum of squares:
\[ \frac1{T^3}\sum_{t=1}^{T}(t-1)^2 =\frac{(T-1)(2T-1)}{6T^2} \longrightarrow\frac13. \]The constant’s diagonal entry is exactly
\[ \frac1T\sum_{t=1}^{T}1^2=1. \]The entry joining the trend and constant is
\[ \frac1{T^2}\sum_{t=1}^{T}(t-1) =\frac{T-1}{2T}\longrightarrow\frac12. \]For \(k=1\), the regressor order is \(y_{t-1},x_{t-1}^+,x_{t-1}^-,q_t,1,z_{2t}’\). The trend entry is therefore \(A_{44,T}\), and the constant entry is \(A_{55,T}\). These are diagonal entries of the normalized cross-product matrix \(A_T\).
With \(k=1\), the drifting block has \(d=4\) rows and columns, and the complete matrix is \((s+5)\times(s+5)\). For the worked lag choice \(p=2\), \(q=1\), there are \(s=3\) short-run regressors, so the complete matrix is \(8\times8\). Its \((4,4)\) and \((5,5)\) positions identify the trend and constant regardless of the sample size.
Two drifting regressors with leading coefficients \(a_i\) and \(a_j\) produce the limit \(a_i a_j/3\). A drifting regressor and the constant produce \(a_i/2\). The weighted sample means of the stationary regressors give the trend–short-run block \(h’/2\), and hence the general block \(ah’/2\).
C.3. Show why the stochastic variation disappears at this scale
For an illustrative scalar drifting regressor \(v_{t-1}=\mu q_t+m_{t-1}\), expand its square:
\[ \frac1{T^3}\sum_t v_{t-1}^2 = \mu^2\frac{\sum_t q_t^2}{T^3} +2\mu\frac{\sum_t q_tm_{t-1}}{T^3} +\frac{\sum_t m_{t-1}^2}{T^3}. \]The last sum has order \(T^2/T^3=T^{-1}\). For the middle sum, the Cauchy–Schwarz inequality gives
\[ \left|\sum_t q_tm_{t-1}\right| \leq \left(\sum_t q_t^2\right)^{1/2} \left(\sum_t m_{t-1}^2\right)^{1/2} =O_p(T^{3/2}T)=O_p(T^{5/2}). \]After division by \(T^3\), that term is \(O_p(T^{-1/2})\). Both stochastic contributions vanish, leaving \(\mu^2/3\). The stochastic component remains in the observations; this normalization suppresses its contribution to the limit.
In probability language, each of the last two normalized terms becomes smaller than any fixed positive tolerance with probability approaching one. Their disappearance is relative to the \(T^3\) normalization. At the finer \(T^2\) scale, the cumulative fluctuations’ sum of squares retains a random limit. The FWL argument uses that finer scale after removing the deterministic trend, which preserves the information lost in the first limiting matrix.
C.4. Establish dependence across the whole matrix
The time-trend position within \(g_t\) is \(2k+2\), and its entry in \(a\) equals one. Every other drifting-regressor column of \(M\) is its corresponding \(a_j\) times that time-trend column. This statement holds in all three row blocks:
\[ \begin{pmatrix} aa_j/3\\ a_j/2\\ ha_j/2 \end{pmatrix} = a_j \begin{pmatrix} a/3\\ 1/2\\ h/2 \end{pmatrix}. \]Thus the full limiting matrix is singular. This explicit column calculation shows how the dependence extends through the constant and short-run rows, so its location in the complete OLS matrix is visible.
Each displayed stacked column has \(d+1+s=K\) entries: a \(d\times1\) upper block, a scalar middle block, and an \(s\times1\) lower block. The dependence therefore concerns full columns of \(M\). The drifting \(d\times d\) block has rank one; the complete matrix can have additional independent columns from the constant and short-run regressors, while remaining singular because its drifting columns are dependent.
C.5. A finite inverse can coexist with a singular limit
A \(2\times2\) matrix illustrates the distinction. Its dimension is fixed, while its entries depend on the sample-size index \(T\):
\[ B_T= \begin{pmatrix}1&1\\1&1+1/T\end{pmatrix}, \qquad \det(B_T)=1\left(1+\frac1T\right)-1\times1=\frac1T>0. \]For every finite positive \(T\), the inverse exists:
\[ B_T^{-1}= \begin{pmatrix}T+1&-T\\-T&T\end{pmatrix}. \]Yet \(B_T\) converges to a matrix with two identical columns. Its diagonal entries stay near one while its determinant tends to zero. The original regression can likewise be estimable in finite samples even though its normalized matrix limit is singular.
The normalized OLS error identity is
Equation (C.3) is the normalized version of Equation (11). The coefficient-error vector has dimension \(K\times1\). Since \(e_t\) is scalar, each \(z_te_t\), and hence their sum, also has dimension \(K\times1\). The matrices \(D\), \(D^{-1}\), and \(A_T^{-1}\) are \(K\times K\), so both sides of the error identity are \(K\times1\) vectors. Each matrix multiplication combines all coefficient directions without changing the number of coefficients.
Since \(M\) has no inverse, the usual step \(A_T^{-1}\to M^{-1}\) is unavailable. This identifies the obstacle in the proof. The transformations in the next appendices retain the stochastic variation at scales that permit a nonsingular limiting matrix.
Main takeaway. The drifting-regressor block has a rank-one limit, and the corresponding full columns remain dependent after the constant and short-run controls are included. A finite-sample inverse can therefore exist even though the normalized matrix has a singular limit.
Return to the main text: End of Section 4: the singular limiting matrix · End of Section 1: motivation · Contents
Appendix D. Derive the first Frisch–Waugh–Lovell transformation
Return to the main text: End of Section 5: the FWL transformation · End of Section 1: motivation · Contents.
D.1. Construct an exact fitted-value and residual decomposition
Regress \(y_{t-1}\) on \(\ddot x_{t-1},q_t,1\), using the same estimation sample as the dynamic regression. Write its regressors and estimated coefficients as columns:
\[ r_{t-1}= \begin{pmatrix} x_{t-1}^{+}\\ x_{t-1}^{-}\\ q_t\\ 1 \end{pmatrix}, \qquad \widetilde\upsilon_T= \begin{pmatrix} \widetilde\beta_T^{+}\\ \widetilde\beta_T^{-}\\ \widetilde\zeta_T\\ \widetilde\nu_T \end{pmatrix}. \]The first two blocks of each column have \(k\) entries each: one for every explanatory variable’s positive partial sum and one for every negative partial sum. The trend contributes one entry, and the constant contributes one entry. Each column therefore contains \(k+k+1+1=2k+2\) entries. In compact notation, \(r_{t-1}=(\ddot x_{t-1}’,q_t,1)’\) and \(\widetilde\upsilon_T=(\widetilde\beta_T’,\widetilde\zeta_T,\widetilde\nu_T)’\).
The symbol \(\widetilde\upsilon_T\) is the estimated coefficient vector, using the Greek letter upsilon. Its capital subscript \(T\) indicates estimation from the sample of \(T\) observations. The number of entries is determined by the regressors, so its size is \((2k+2)\times1\). Define the fitted residual by
Here \(r_{t-1}\) denotes the auxiliary regressor vector; Section 4 used \(r_t\) for departures from deterministic trends. The product \(r_{t-1}’\widetilde\upsilon_T\) multiplies a \(1\times(2k+2)\) row by a \((2k+2)\times1\) column, giving one scalar fitted value. For \(k=1\), all four entries can be seen directly:
\[ \begin{aligned} r_{t-1}’\widetilde\upsilon_T &=\begin{pmatrix}x_{t-1}^{+}&x_{t-1}^{-}&q_t&1\end{pmatrix} \begin{pmatrix} \widetilde\beta_T^{+}\\ \widetilde\beta_T^{-}\\ \widetilde\zeta_T\\ \widetilde\nu_T \end{pmatrix}\\ &=x_{t-1}^{+}\widetilde\beta_T^{+} +x_{t-1}^{-}\widetilde\beta_T^{-} +q_t\widetilde\zeta_T+\widetilde\nu_T. \end{aligned} \]Each multiplication pairs one observed regressor with its estimated coefficient; adding the four products gives the fitted lagged outcome. Stacking the auxiliary regressor rows across dates gives a \(T\times(2k+2)\) design matrix.
The definition gives the exact sample identity
\[ y_{t-1} =\widetilde\beta_T’\ddot x_{t-1} +\widetilde\zeta_Tq_t+\widetilde\nu_T +\widetilde u_{t-1}. \]OLS implies \(\sum_t r_{t-1}\widetilde u_{t-1}=0\): the residual is orthogonal to each regressor used in this projection. The short-run variables \(z_{2t}\) were not part of that projection, so \(\sum_tz_{2t}\widetilde u_{t-1}\) generally need not be zero.
The first sum is a \((2k+2)\times1\) vector, and its zero is a zero vector of that dimension: one orthogonality condition for each auxiliary regressor. The second sum has dimension \(s\times1\). The estimated residual \(\widetilde u_{t-1}\) itself is scalar at each date.
D.2. Retain the other columns in the dynamic regression
Now regress \(\Delta y_t\) on \(\widetilde u_{t-1},\ddot x_{t-1},q_t,1,z_{2t}\). Its fitted component is
\[ \widetilde\rho_T\widetilde u_{t-1} +\widetilde\eta_T’\ddot x_{t-1} +\widetilde\psi_Tq_t+\widetilde\gamma_T +\widetilde\kappa_T’z_{2t}. \]Substitute the residual’s definition from Equation (D.1) into the first term:
\[ \begin{aligned} \widetilde\rho_T\widetilde u_{t-1} ={}&\widetilde\rho_Ty_{t-1} -\widetilde\rho_T\widetilde\beta_T’\ddot x_{t-1}\\ &-\widetilde\rho_T\widetilde\zeta_Tq_t -\widetilde\rho_T\widetilde\nu_T. \end{aligned} \]Now add back the other terms of the fitted dynamic equation and group those multiplying the same original regressor:
\[ \begin{aligned} &\widetilde\rho_T\widetilde u_{t-1} +\widetilde\eta_T’\ddot x_{t-1} +\widetilde\psi_Tq_t+\widetilde\gamma_T +\widetilde\kappa_T’z_{2t}\\ &\quad=\widetilde\rho_Ty_{t-1} +(\widetilde\eta_T-\widetilde\rho_T\widetilde\beta_T)’\ddot x_{t-1} +(\widetilde\psi_T-\widetilde\rho_T\widetilde\zeta_T)q_t\\ &\qquad\quad+(\widetilde\gamma_T-\widetilde\rho_T\widetilde\nu_T) +\widetilde\kappa_T’z_{2t}. \end{aligned} \]For example, the original partial-sum column appears twice after substitution: directly through \(\widetilde\eta_T\), and inside the generated residual through \(-\widetilde\rho_T\widetilde\beta_T\). Adding those contributions gives its original-regression coefficient. Reading every column in this way gives
Each minus sign in Equation (D.2) comes from subtracting a fitted component when constructing the residual. For an arithmetic illustration, if a component of \(\widetilde\beta_T\) equals \(2\), \(\widetilde\rho_T=-0.4\), and the corresponding \(\widetilde\eta_T=0.03\), then the reconstructed coefficient is \(0.03-(-0.4)(2)=0.83\). These numbers illustrate the identity; they are not empirical estimates.
D.3. Verify equality of the OLS estimators using matrices
Construct the original design matrix first. For one observation, place every regressor in the order used in the original dynamic equation:
\[ z_t’=\begin{pmatrix} y_{t-1}&\ddot x_{t-1}’&q_t&1&z_{2t}’ \end{pmatrix} =\begin{pmatrix}y_{t-1}&r_{t-1}’&z_{2t}’\end{pmatrix}. \]The expanded row contains \(1+2k+1+1+s=K\) entries. The grouped version has the same entries in three blocks: the lagged outcome, the \(2k+2\) auxiliary regressors, and the \(s\) short-run regressors. Stacking these rows for \(t=1,\ldots,T\) defines \(Z\):
\[ Z=\begin{pmatrix}z_1’\\z_2’\\\vdots\\z_T’\end{pmatrix} =\begin{pmatrix} y_0&r_0’&z_{2,1}’\\ y_1&r_1’&z_{2,2}’\\ \vdots&\vdots&\vdots\\ y_{T-1}&r_{T-1}’&z_{2,T}’ \end{pmatrix}, \qquad Y=\begin{pmatrix}\Delta y_1\\\Delta y_2\\\vdots\\\Delta y_T\end{pmatrix}. \]Each row of \(Z\) corresponds to one observation; each column corresponds to one regressor. Thus \(Z\) is \(T\times K\), whereas the outcome column \(Y\) is \(T\times1\). The subscript in \(z_{2,t}\) identifies the short-run regressor block at date \(t\).
To replace the first column by the auxiliary residual, while retaining the other columns, multiply \(Z\) on the right by
\[ R_T= \begin{pmatrix} 1&0&0\\ -\widetilde\upsilon_T&I_{2k+2}&0\\ 0&0&I_s \end{pmatrix}. \]The three block row heights and column widths of \(R_T\) are \(1\), \(2k+2\), and \(s\), matching the three groups in \(Z\). Their sum is \(K\), so \(R_T\) is \(K\times K\). In the first column, \(-\widetilde\upsilon_T\) occupies \(2k+2\) rows because it contains the negative of every auxiliary coefficient. When \(k=1\), this block is explicitly
\[ -\widetilde\upsilon_T= \begin{pmatrix} -\widetilde\beta_T^{+}\\ -\widetilde\beta_T^{-}\\ -\widetilde\zeta_T\\ -\widetilde\nu_T \end{pmatrix}, \qquad\text{a }4\times1\text{ column}. \]If there are also \(s=3\) short-run regressors, then \(K=1+4+3=8\): \(R_T\) is \(8\times8\), with diagonal blocks \(1\), \(I_4\), and \(I_3\). For arbitrary \(k\) and \(s\), the zeros in its first row are \(1\times(2k+2)\) and \(1\times s\) blocks. Each remaining zero block takes the row height and column width required by its position.
Compute one row of \(ZR_T\) to see how the transformation works:
\[ \begin{aligned} &\begin{pmatrix}y_{t-1}&r_{t-1}’&z_{2t}’\end{pmatrix} \begin{pmatrix} 1&0&0\\ -\widetilde\upsilon_T&I_{2k+2}&0\\ 0&0&I_s \end{pmatrix}\\ &\qquad= \begin{pmatrix} y_{t-1}-r_{t-1}’\widetilde\upsilon_T&r_{t-1}’&z_{2t}’ \end{pmatrix}\\ &\qquad= \begin{pmatrix}\widetilde u_{t-1}&r_{t-1}’&z_{2t}’\end{pmatrix}. \end{aligned} \]The first entry is the row multiplied by the first column: \(y_{t-1}\times1+r_{t-1}'(-\widetilde\upsilon_T)+z_{2t}’0\). This subtracts the fitted long-run value from the lagged outcome. The identity blocks copy the remaining entries. Repeating the operation for all \(T\) rows gives
\[ \widetilde Z=ZR_T= \begin{pmatrix} \widetilde u_0&r_0’&z_{2,1}’\\ \widetilde u_1&r_1’&z_{2,2}’\\ \vdots&\vdots&\vdots\\ \widetilde u_{T-1}&r_{T-1}’&z_{2,T}’ \end{pmatrix}. \]Compare this display with \(Z\): its first column changes from \(y_{t-1}\) to \(\widetilde u_{t-1}=y_{t-1}-r_{t-1}’\widetilde\upsilon_T\). All remaining columns retain their original observations. The multiplication \((T\times K)(K\times K)\) therefore produces another \(T\times K\) design matrix.
Its coefficient vector \(\widetilde\omega_T=(\widetilde\rho_T,\widetilde\eta_T’,\widetilde\psi_T,\widetilde\gamma_T,\widetilde\kappa_T’)’\) is \(K\times1\). The matrix \(R_T\) is triangular with ones on its diagonal, so it is invertible and has determinant one. Under full sample rank, apply the usual OLS formula:
\[ \begin{aligned} \widetilde\omega_T &=(\widetilde Z’\widetilde Z)^{-1}\widetilde Z’Y\\ &=(R_T’Z’ZR_T)^{-1}R_T’Z’Y. \end{aligned} \]To simplify the inverse, reverse the order of the factors and invert each one. Treating \(Z’Z\) as one factor gives
\[ (R_T’Z’ZR_T)^{-1} =R_T^{-1}(Z’Z)^{-1}(R_T’)^{-1}. \]The order matters because matrix multiplication generally cannot be rearranged. Substitute this expression into the OLS formula; the adjacent factors \((R_T’)^{-1}R_T’\) then cancel:
\[ \begin{aligned} \widetilde\omega_T &=R_T^{-1}(Z’Z)^{-1} \underbrace{(R_T’)^{-1}R_T’}_{I_K}Z’Y\\ &=R_T^{-1}(Z’Z)^{-1}Z’Y\\ &=R_T^{-1}\widehat{\boldsymbol\alpha}_T. \end{aligned} \]All three factors in the inverse identity are \(K\times K\). The vector \(Z’Y\) is \(K\times1\), so the resulting coefficient vector has one entry for each of the \(K\) regressors.
Multiplying by \(R_T\) produces Proposition 1:
Equation (D.3) collects the coefficient identities in Equation (D.2) into one matrix multiplication. The fitted values also coincide: \(\widetilde Z\widetilde\omega_T=ZR_TR_T^{-1}\widehat{\boldsymbol\alpha}_T=Z\widehat{\boldsymbol\alpha}_T\). Their dynamic regression residuals are therefore identical.
Both fitted-value vectors have dimension \(T\times1\): a \(T\times K\) design matrix multiplies a \(K\times1\) coefficient vector to produce one fitted value for every observation. The transformation changes the coordinate system for the coefficients while retaining the same number of regressors and fitted observations.
The population coefficients on the extra levels and trend would be zero if the true cointegration error were used. With an estimated residual, those columns are retained to preserve the exact sample equivalence. The representation reconstructs every original coefficient and supplies the algebra needed for the asymptotic analysis.
Appendices A–D unpack equations (1)–(7), Lemma 1, and Proposition 1 in Cho’s paper. The small numerical and matrix examples are explanatory constructions.
Main takeaway. Replacing the lagged outcome by its auxiliary residual, while retaining the other regressors, is an invertible change of columns. The fitted values stay the same, and the original coefficients follow from an exact matrix identity.
Return to the main text: End of Section 5: the FWL transformation · End of Section 1: motivation · Contents
Appendix E. Propositions 2–4: remove the fitted trends, then recover every OLS coefficient
Return to the main text: End of Section 5: detrending · End of Section 6: recovering the original coefficients · Contents.
E.1. A change of regressors that preserves the regression
Let \(q_t=t-1\), a shorthand for the paper’s lagged time trend, and let \(\ddot x_{t-1}\) stack the \(2k\) positive and negative partial sums. Double dots are notation labels, not derivatives; on \(\ddot x_{t-1}\), they identify the stacked partial sums. Cho first projects each partial sum on \(q_t\), without an intercept:
The definitions in Equation (E.1) repeat Equation (13) and give two exact sample identities:
\[ \ddot x_{t-1}=\ddot m_{t-1}+q_t\ddot\mu_T, \qquad \sum_{t=1}^{T}q_t\ddot m_{t-1}=0. \]Dimensions. At one date, \(\ddot x_{t-1}\), \(\ddot m_{t-1}\), and \(\ddot\mu_T\) are \(2k\times1\) column vectors. The numerator in the drift estimate is a sum of \(2k\times1\) vectors, while its denominator is a scalar; the resulting estimate therefore has one entry for each partial sum. The zero in the orthogonality identity above is a \(2k\times1\) zero vector, giving one orthogonality condition per partial sum.
The first reconstructs the original partial sums. The second is the OLS orthogonality condition: the residuals have zero sample cross-product with the fitted trend. Their sample means generally remain nonzero, because this projection has no intercept. The subsequent regressions therefore retain a constant.
The original auxiliary long-run regression uses \((\ddot x_{t-1}’,q_t,1)’\), with coefficient vector \(\widetilde\upsilon_T=(\widetilde\beta_T’,\widetilde\zeta_T,\widetilde\nu_T)’\). Replacing the partial sums by their detrended counterparts gives a regression on \((\ddot m_{t-1}’,q_t,1)’\), with coefficients \(\ddot\upsilon_T=(\ddot\beta_T’,\ddot\vartheta_T,\ddot\nu_T)’\). Start with its fitted decomposition and substitute the definition of \(\ddot m_{t-1}\):
\[ \begin{aligned} y_{t-1} &=\ddot\beta_T’\ddot m_{t-1}+\ddot\vartheta_Tq_t+\ddot\nu_T+\ddot u_{t-1}\\ &=\ddot\beta_T'(\ddot x_{t-1}-q_t\ddot\mu_T) +\ddot\vartheta_Tq_t+\ddot\nu_T+\ddot u_{t-1}\\ &=\ddot\beta_T’\ddot x_{t-1} -\ddot\beta_T’\ddot\mu_Tq_t +\ddot\vartheta_Tq_t+\ddot\nu_T+\ddot u_{t-1}\\ &=\ddot\beta_T’\ddot x_{t-1} +(\ddot\vartheta_T-\ddot\mu_T’\ddot\beta_T)q_t +\ddot\nu_T+\ddot u_{t-1}. \end{aligned} \]The last line groups the two terms multiplying \(q_t\). The products \(\ddot\beta_T’\ddot\mu_T\) and \(\ddot\mu_T’\ddot\beta_T\) are the same scalar sum of matching entries. Matching coefficients on the original columns gives Proposition 2:
In Equation (E.2), the transformed trend coefficient absorbs the trends previously carried inside the partial sums. Recovering the original explicit trend requires subtracting their fitted contribution. The intercept stays unchanged because the decomposition of \(\ddot x_{t-1}\) adds only a multiple of \(q_t\).
The two auxiliary coefficient vectors \(\widetilde\upsilon_T\) and \(\ddot\upsilon_T\) each have \(2k+2\) entries: \(2k\) partial-sum slopes, one trend coefficient, and one intercept. Their regression matrices have \(T\) rows and \(2k+2\) columns. In the trend correction, \(\ddot\mu_T’\) is \(1\times2k\) and \(\ddot\beta_T\) is \(2k\times1\), so their product is the scalar needed to adjust a single trend coefficient. For \(k=1\), the transformation below is a \(4\times4\) matrix multiplying a \(4\times1\) coefficient vector.
For one explanatory variable, the matrix form makes every adjustment visible:
\[ \begin{pmatrix}\widetilde\beta_T^+\\\widetilde\beta_T^-\\\widetilde\zeta_T\\\widetilde\nu_T\end{pmatrix} = \underbrace{\begin{pmatrix} 1&0&0&0\\ 0&1&0&0\\ -\ddot\mu_T^+&-\ddot\mu_T^-&1&0\\ 0&0&0&1 \end{pmatrix}}_{\mathbf P_T} \begin{pmatrix}\ddot\beta_T^+\\\ddot\beta_T^-\\\ddot\vartheta_T\\\ddot\nu_T\end{pmatrix}. \]For example, take \(\ddot\mu_T=(0.4,-0.4)’\), \(\ddot\beta_T=(2,1)’\), and \(\ddot\vartheta_T=0.7\). The fitted trend contributed by the partial sums is \(0.4(2)-0.4(1)=0.4\), giving \(\widetilde\zeta_T=0.7-0.4=0.3\).
Both regressor sets span the same sample column space. With full column rank, their fitted values are identical, so the estimated cointegration residual is also identical: \(\widetilde u_{t-1}=\ddot u_{t-1}\). Detrending changes the coordinates used to describe the fitted relationship.
This equivalence can be checked in both directions. Starting with the original regression, replace each partial sum by its detrended component plus fitted trend. Starting with the transformed regression, add the fitted trend back into each partial sum. Every fitted observation is reproduced. The coefficient on a detrended partial sum remains the original slope because the coefficient multiplying that component has not changed; the trend coefficient accounts for the contribution moved into the time regressor.
E.2. Apply the same transformation to the dynamic regression
Proposition 3 repeats this operation with \(\Delta y_t\) as the dependent variable. The fully augmented transformed regression is
\[ \Delta y_t=\ddot\rho_T\ddot u_{t-1} +\ddot\eta_T’\ddot m_{t-1}+\ddot\varsigma_Tq_t +\ddot\gamma_T+\ddot\kappa_T’z_{2t}+\ddot e_t. \]Here, \(z_{2t}\) contains the \(s=p+2kq-1\) short-run regressors. Distinguish \(\ddot\vartheta_T\), the trend coefficient in the auxiliary long-run regression, from \(\ddot\varsigma_T\), the trend coefficient in this dynamic regression. They belong to different equations.
For dimension counting, write \(K:=1+2k+1+1+s=2k+s+3\) for the total number of coefficients in the dynamic regression. The five blocks correspond to the adjustment coefficient, the \(2k\) detrended partial sums, the trend, the intercept, and the \(s\) short-run regressors. Thus, \(\ddot\eta_T\) and \(\ddot m_{t-1}\) are \(2k\times1\), while \(\ddot\kappa_T\) and \(z_{2t}\) are \(s\times1\). Both inner products in the equation are scalars: \((1\times2k)(2k\times1)=1\times1\) and \((1\times s)(s\times1)=1\times1\).
In this dynamic equation, substitute \(\ddot m_{t-1}=\ddot x_{t-1}-q_t\ddot\mu_T\) and distribute the partial-sum coefficients:
\[ \begin{aligned} \Delta y_t &=\ddot\rho_T\ddot u_{t-1} +\ddot\eta_T'(\ddot x_{t-1}-q_t\ddot\mu_T) +\ddot\varsigma_Tq_t+\ddot\gamma_T +\ddot\kappa_T’z_{2t}+\ddot e_t\\ &=\ddot\rho_T\ddot u_{t-1} +\ddot\eta_T’\ddot x_{t-1} -\ddot\eta_T’\ddot\mu_Tq_t +\ddot\varsigma_Tq_t+\ddot\gamma_T +\ddot\kappa_T’z_{2t}+\ddot e_t\\ &=\ddot\rho_T\ddot u_{t-1} +\ddot\eta_T’\ddot x_{t-1} +(\ddot\varsigma_T-\ddot\mu_T’\ddot\eta_T)q_t +\ddot\gamma_T+\ddot\kappa_T’z_{2t}+\ddot e_t. \end{aligned} \]The subtraction generated by detrending joins the original trend term. Because \(\ddot u_{t-1}=\widetilde u_{t-1}\), every regressor in the final line is now a column of the first transformed dynamic regression. Matching those columns gives
\[ \widetilde\rho_T=\ddot\rho_T,\quad \widetilde\eta_T=\ddot\eta_T,\quad \widetilde\psi_T=\ddot\varsigma_T-\ddot\mu_T’\ddot\eta_T,\quad \widetilde\gamma_T=\ddot\gamma_T,\quad \widetilde\kappa_T=\ddot\kappa_T. \]Only the explicit trend coefficient changes. The correction now uses \(\ddot\eta_T\), the coefficients from the dynamic regression. Writing these identities as \(\widetilde\omega_T=\mathbf Q_T\ddot\tau_T\), the matrix \(\mathbf Q_T\) is an identity matrix with the additional entries \(-\ddot\mu_T’\) in the trend row and the \(\ddot\eta_T\) columns.
\[ \mathbf Q_T=\begin{pmatrix}I_{1+2k}&0\\\mathbf Q_T^{21}&I_{s+2}\end{pmatrix}, \qquad \mathbf Q_T^{21}=\begin{pmatrix}0&-\ddot\mu_T’\\0_{(s+1)\times1}&0_{(s+1)\times2k}\end{pmatrix}. \]The first block contains the adjustment coefficient and the partial-sum coefficients. The lower block contains the trend, constant, and short-run coefficients. Its leading zero means that the trend correction contains no adjustment-coefficient term at this step: the cointegration residual itself remains unchanged.
The matrix \(\mathbf Q_T\) is \(K\times K\). Its upper block has \(1+2k\) rows and its lower block has \(s+2\) rows. Consequently, \(\mathbf Q_T^{21}\) is \((s+2)\times(1+2k)\), and the upper-right zero block is \((1+2k)\times(s+2)\). Within \(\mathbf Q_T^{21}\), the first row contains one scalar zero followed by the \(1\times2k\) row \(-\ddot\mu_T’\). The remaining \(s+1\) rows are zeros. This locates the correction in the trend equation while leaving the intercept and short-run coefficients unchanged.
E.3. Reconstruct the original one-step estimator
Proposition 1 already gives \(\widehat{\boldsymbol\alpha}_T=\mathbf R_T\widetilde\omega_T\), using the identity defining the estimated cointegration residual. Combining the transformations gives Proposition 4:
\[ \mathbf R_T=\operatorname{diag}\left[ \begin{pmatrix}1&0_{1\times(2k+2)}\\-\widetilde\upsilon_T&I_{2k+2}\end{pmatrix},I_s\right]. \]The negative signs come directly from \(\widetilde u_{t-1}=y_{t-1}-\widetilde\beta_T’\ddot x_{t-1}-\widetilde\zeta_Tq_t-\widetilde\nu_T\). Multiplying this identity by the estimated adjustment coefficient distributes its contribution across the lagged dependent variable, partial sums, trend, and constant.
The vector \(\widetilde\upsilon_T\) is \((2k+2)\times1\). The first diagonal block of \(\mathbf R_T\) therefore has \(2k+3\) rows and columns: one for the adjustment coefficient and \(2k+2\) for the auxiliary coefficients. Appending the \(s\times s\) identity block gives a \(K\times K\) matrix. Thus \(\mathbf R_T\), \(\mathbf Q_T\), and their product \(\mathbf T_T\) are all \(K\times K\); each maps one \(K\times1\) coefficient vector into another.
The transformations in Equation (E.3) act from right to left. The bold matrix \(\mathbf T_T\) depends on the sample; it is distinct from the scalar sample size \(T\). For \(k=1\), the full matrix has size \((s+5)\times(s+5)\). Its leading \(5\times5\) block maps the five coefficients shown below; the omitted columns for the short-run coefficients are zeros in these five rows:
\[ \begin{pmatrix}\widehat\rho_T\\\widehat\theta_T^+\\\widehat\theta_T^-\\\widehat\xi_T\\\widehat\alpha_T\end{pmatrix} = \begin{pmatrix} 1&0&0&0&0\\ -\ddot\beta_T^+&1&0&0&0\\ -\ddot\beta_T^-&0&1&0&0\\ -\widetilde\zeta_T&-\ddot\mu_T^+&-\ddot\mu_T^-&1&0\\ -\ddot\nu_T&0&0&0&1 \end{pmatrix} \begin{pmatrix}\ddot\rho_T\\\ddot\eta_T^+\\\ddot\eta_T^-\\\ddot\varsigma_T\\\ddot\gamma_T\end{pmatrix}. \]Each row gives one scalar coefficient. For the positive partial sum, multiply the second row by the coefficient column:
\[ \begin{aligned} \widehat\theta_T^+ &=(-\ddot\beta_T^+)\ddot\rho_T +1\times\ddot\eta_T^+ +0\times\ddot\eta_T^- +0\times\ddot\varsigma_T +0\times\ddot\gamma_T\\ &=\ddot\eta_T^+-\ddot\rho_T\ddot\beta_T^+. \end{aligned} \]The fourth row recovers the trend coefficient and displays both drift corrections:
\[ \begin{aligned} \widehat\xi_T &=(-\widetilde\zeta_T)\ddot\rho_T +(-\ddot\mu_T^+)\ddot\eta_T^+ +(-\ddot\mu_T^-)\ddot\eta_T^- +\ddot\varsigma_T\\ &=\ddot\varsigma_T-\ddot\rho_T\widetilde\zeta_T -\bigl(\ddot\mu_T^+\ddot\eta_T^+ +\ddot\mu_T^-\ddot\eta_T^-\bigr)\\ &=\ddot\varsigma_T-\ddot\rho_T\widetilde\zeta_T -\ddot\mu_T’\ddot\eta_T. \end{aligned} \]The identity block for the short-run coefficients simply copies those coefficients. Reading every row, now allowing any number \(k\) of explanatory variables, gives
Equation (E.2) supplies \(\widetilde\zeta_T=\ddot\vartheta_T-\ddot\mu_T’\ddot\beta_T\). Substituting it into the trend row of Equation (E.4) gives an equivalent formula:
\[ \begin{aligned} \widehat\xi_T &=\ddot\varsigma_T -\ddot\rho_T(\ddot\vartheta_T-\ddot\mu_T’\ddot\beta_T) -\ddot\mu_T’\ddot\eta_T\\ &=\ddot\varsigma_T-\ddot\rho_T\ddot\vartheta_T -\ddot\mu_T'(\ddot\eta_T-\ddot\rho_T\ddot\beta_T)\\ &=\ddot\varsigma_T-\ddot\rho_T\ddot\vartheta_T -\ddot\mu_T’\widehat\theta_T. \end{aligned} \]The second line factors out \(-\ddot\mu_T’\); the last uses the slope reconstruction already derived. These are exact finite-sample identities wherever the required regressions have full rank. They retain all regressors and their jointly estimated coefficients. Their purpose is to expose the components whose probability limits must now be studied.
Each transformation matrix is triangular with ones on its diagonal, so its determinant is one. This establishes a one-to-one correspondence between coefficient vectors. The auxiliary residual is orthogonal to the regressors in the long-run regression, but it can remain correlated with the short-run regressors. Keeping those regressors in the dynamic equation preserves the complete original estimation problem.
Main takeaway. Detrending transfers the fitted drift contributions into the explicit trend coefficient. Combining this transformation with the first FWL identity recovers every original coefficient, including the intercept and short-run coefficients.
Source: Cho, Section 3, Propositions 2–4. Return to the main text: End of Section 5: detrending · End of Section 6: recovering the original coefficients · Contents.
Appendix F. Lemmas 2–3: nonsingular limits and the long-run estimator
Return to the main text: End of Section 5: the transformed matrix · End of Section 7: probability limits · Contents.
F.1. Preserve the stochastic fluctuations at their own scale
First distinguish the centered partial sum from the residual obtained by fitting a trend. The centered process is \(m_{t-1}=\ddot x_{t-1}-\mu_*q_t\), whereas the detrended residual uses the estimated drift \(\ddot\mu_T\). Substituting the first definition into the second gives
\[ \begin{aligned} \ddot m_{t-1} &=\ddot x_{t-1}-q_t\ddot\mu_T\\ &=\mu_*q_t+m_{t-1}-q_t\ddot\mu_T\\ &=m_{t-1}-q_t(\ddot\mu_T-\mu_*). \end{aligned} \]The fitted residual therefore includes the original cumulative fluctuations and a correction for the error in estimating the drift. That correction will remain visible in the limiting calculation.
Place samples of different lengths on the same time axis. At observation \(t\), write \(r_{t,T}=(t-1)/T=q_t/T\). The grid is \(0,1/T,\ldots,(T-1)/T\), with spacing \(1/T\). Thus a point near \(r=1/2\) refers to the middle of each sample, whether that sample contains 100 or 400 observations. Dividing a centered partial sum by \(\sqrt T\) also puts its stochastic magnitude on a fixed scale. Define the path
\[ \mathcal B_{s,T}(r):=\frac{m_{\lfloor Tr\rfloor}}{\sqrt T}, \qquad 0\leq r\leq1. \]The symbol \(\lfloor Tr\rfloor\) selects the integer observation index at or just before the sample fraction \(r\). The paper’s functional limit theorem describes the distributional limit of this entire rescaled path by \(\mathcal B_s(r)\). It is a \(2k\times1\) Brownian vector whose covariance reflects the underlying centered increments; its components need not be independent or have unit variance. The subscript \(s\) refers to those centered increments.
Recover the factor three from the fitted trend. The trend projection gives an exact expression for its slope error:
\[ \begin{aligned} \ddot\mu_T-\mu_* &=\frac{\sum_{t=1}^{T}q_tm_{t-1}}{\sum_{t=1}^{T}q_t^2},\\ \sqrt T(\ddot\mu_T-\mu_*) &=\frac{\displaystyle\frac1T\sum_{t=1}^{T} \left(\frac{q_t}{T}\right)\left(\frac{m_{t-1}}{\sqrt T}\right)} {\displaystyle\frac1T\sum_{t=1}^{T}\left(\frac{q_t}{T}\right)^2}. \end{aligned} \]In the numerator, each scaled cumulative observation is weighted by its position \(r_{t,T}\) in the sample. The factor \(1/T\) is the width of each interval on the unit time axis. Under the functional limit assumptions, these sums approach integrals of the limiting path. The denominator is an ordinary sum of squared time fractions:
\[ \sqrt T(\ddot\mu_T-\mu_*) \Rightarrow \frac{\displaystyle\int_0^1r\mathcal B_s(r)\,dr} {\displaystyle\int_0^1r^2\,dr} =3\int_0^1r\mathcal B_s(r)\,dr, \qquad \int_0^1r^2\,dr=\frac13. \]This explains the factor three and the \(T^{-1/2}\) size of the drift-estimation error. Now divide the detrended-residual identity by \(\sqrt T\):
\[ \frac{\ddot m_{t-1}}{\sqrt T} =\frac{m_{t-1}}{\sqrt T} -\frac{q_t}{T}\left[\sqrt T(\ddot\mu_T-\mu_*)\right]. \]The scaled cumulative process and the scaled drift-estimation error remain on finite stochastic scales. At a sample fraction \(r\), the resulting limiting path is
In Equation (F.1), the integration variable \(v\) runs over the whole unit interval; \(r\) is the point at which we evaluate the detrended path. Each integral is taken component by component. The projection’s orthogonality can now be checked directly:
\[ \begin{aligned} \int_0^1r\overline{\mathcal B}_s(r)\,dr &=\int_0^1r\mathcal B_s(r)\,dr -3\left(\int_0^1r^2\,dr\right) \left(\int_0^1v\mathcal B_s(v)\,dv\right)\\ &=\int_0^1r\mathcal B_s(r)\,dr -\int_0^1v\mathcal B_s(v)\,dv =0_{2k}. \end{aligned} \]The last two integrals are the same vector: changing the name of their integration variable does not change their value. This is the limiting counterpart of \(\sum_tq_t\ddot m_{t-1}=0\).
From one path to a whole regression column. Dividing an observation by \(\sqrt T\) makes its path comparable across sample sizes. The cross-product matrix also adds contributions from all \(T\) observations. Its detrended-partial-sum block is therefore normalized as
\[ \begin{aligned} \frac1{T^2}\sum_{t=1}^{T}\ddot m_{t-1}\ddot m_{t-1}^{\prime} &=\frac1T\sum_{t=1}^{T} \left(\frac{\ddot m_{t-1}}{\sqrt T}\right) \left(\frac{\ddot m_{t-1}}{\sqrt T}\right)^{\prime}\\ &\Rightarrow\int_0^1 \overline{\mathcal B}_s(r)\overline{\mathcal B}_s(r)^{\prime}\,dr. \end{aligned} \]The first line is an exact rearrangement: the two factors \(1/\sqrt T\) contribute \(1/T\), and the averaging factor contributes another \(1/T\). The sum becomes a random integral because its summands follow the random limiting path. Equivalently, dividing the entire regressor column by \(T\) divides each cross-product by \(T^2\). The path divisor \(\sqrt T\) and the column divisor \(T\) serve these two distinct calculations.
For \(k=1\), write the two components of \(\overline{\mathcal B}_s(r)\) as \(B_+(r)\) and \(B_-(r)\) for this display. The product inside the integral is
\[ \begin{pmatrix}B_+(r)\\B_-(r)\end{pmatrix} \begin{pmatrix}B_+(r)&B_-(r)\end{pmatrix} =\begin{pmatrix} B_+(r)^2&B_+(r)B_-(r)\\ B_-(r)B_+(r)&B_-(r)^2 \end{pmatrix}. \]The first diagonal entry accumulates squared positive-component fluctuations. The off-diagonal entry accumulates products of the positive and negative fluctuations at the same sample fraction. Every entry is a scalar integral, just as every entry of a finite cross-product matrix is a scalar sum.
The constant and trend columns can be handled the same way. Their cross-products with the detrended partial sums are
\[ \begin{aligned} \sum_{t=1}^{T}\frac{\ddot m_{t-1}}{T}\frac1{\sqrt T} &=\frac1T\sum_{t=1}^{T}\frac{\ddot m_{t-1}}{\sqrt T} \Rightarrow\int_0^1\overline{\mathcal B}_s(r)\,dr,\\ \sum_{t=1}^{T}\frac{\ddot m_{t-1}}{T}\frac{q_t}{T^{3/2}} &=\frac1T\sum_{t=1}^{T} \frac{\ddot m_{t-1}}{\sqrt T}\frac{q_t}{T} =0_{2k}. \end{aligned} \]The first generally remains random. The second is exactly zero because of the fitted trend projection. A zero cross-product with the trend therefore does not imply a zero cross-product with the constant.
The auxiliary regression contains the \(2k\) detrended partial sums, one trend, and one constant. Its regressor vector and column normalizer are
\[ \ddot r_{t-1}=(\ddot m_{t-1}^{\prime},q_t,1)^{\prime}, \qquad \ddot D_1=\operatorname{diag}(TI_{2k},T^{3/2},T^{1/2}). \]They have dimensions \((2k+2)\times1\) and \((2k+2)\times(2k+2)\), respectively. Assembling the entries just calculated gives the limiting cross-product matrix
The block sizes in Equation (F.2) are \(2k,1,1\). Thus the upper-left block is \(2k\times2k\), while the trend and constant occupy the final two rows and columns. Their entries \(1/3\), \(1/2\), and \(1\) are the same elementary integrals derived in Appendix C.
Why the retained variation permits inversion. Define \(\psi(r)=(\overline{\mathcal B}_s(r)^{\prime},r,1)^{\prime}\). Then \(\mathcal M_{11}=\int_0^1\psi(r)\psi(r)^{\prime}\,dr\); a matrix built from these pairwise products is called a Gram matrix. To see what a linear combination of its columns represents, take a \(2k\times1\) vector \(b\) and scalars \(c,\ell\), and stack them as \(w=(b^{\prime},c,\ell)^{\prime}\). Multiplying the row by the column gives
\[ w^{\prime}\psi(r) =\begin{pmatrix}b^{\prime}&c&\ell\end{pmatrix} \begin{pmatrix}\overline{\mathcal B}_s(r)\\r\\1\end{pmatrix} =b^{\prime}\overline{\mathcal B}_s(r)+cr+\ell. \]Now carry this scalar product into the quadratic form:
\[ \begin{aligned} w^{\prime}\mathcal M_{11}w &=\int_0^1\bigl[w^{\prime}\psi(r)\bigr] \bigl[\psi(r)^{\prime}w\bigr]\,dr\\ &=\int_0^1\bigl[b^{\prime}\overline{\mathcal B}_s(r)+cr+\ell\bigr]^2\,dr\\ &=\int_0^1\bigl[b^{\prime}\overline{\mathcal B}_s(r)\bigr]^2\,dr +\frac{c^2}{3}+\ell^2\\ &\quad+2c\,b^{\prime}\int_0^1r\overline{\mathcal B}_s(r)\,dr +2\ell\,b^{\prime}\int_0^1\overline{\mathcal B}_s(r)\,dr+c\ell. \end{aligned} \]The two bracketed factors in the first line are the same scalar, which explains the square in the second line. The final line uses \(\int r^2dr=1/3\) and \(2\int r\,dr=1\). The term involving \(\int r\overline{\mathcal B}_sdr\) vanishes by the projection calculation above.
An integral of a continuous squared function can equal zero only when the function is zero throughout the interval. Under the paper’s nondegeneracy assumptions, no nonzero choice of \(b,c,\ell\) makes this combination vanish. Lemma 2 consequently establishes that \(\mathcal M_{11}\) is nonsingular with probability one. The full transformed dynamic regression has a larger \(K\times K\) limiting matrix \(\mathcal M\), also nonsingular under the maintained assumptions. The cumulative fluctuations that disappeared in the first normalization now supply independent variation in these matrices.
F.2. Why the trend coefficient has a sample-dependent target
Start with the population long-run relationship and replace each partial sum by its detrended residual plus its fitted trend. The coefficients on the population relationship remain fixed throughout:
\[ \begin{aligned} y_{t-1} &=\beta_*^{\prime}\ddot x_{t-1}+\zeta_*q_t+\nu_*+u_{t-1}\\ &=\beta_*^{\prime}(\ddot m_{t-1}+q_t\ddot\mu_T) +\zeta_*q_t+\nu_*+u_{t-1}\\ &=\beta_*^{\prime}\ddot m_{t-1} +(\zeta_*+\ddot\mu_T^{\prime}\beta_*)q_t+\nu_*+u_{t-1}. \end{aligned} \]The scalar products \(\beta_*^{\prime}\ddot\mu_T\) and \(\ddot\mu_T^{\prime}\beta_*\) are equal. Naming the coefficient now attached to the trend gives
\[ y_{t-1}=\beta_*’\ddot m_{t-1}+\vartheta_{T*}q_t+\nu_*+u_{t-1}, \qquad \vartheta_{T*}=\zeta_*+\ddot\mu_T’\beta_*. \]The target changes with the fitted drift because that drift defines the transformed regressors. Its fixed probability limit is \(\vartheta_*:=\zeta_*+\mu_*’\beta_*\). Define \(\overline\upsilon_{T*}=(\beta_*’,\vartheta_{T*},\nu_*)’\). The OLS error has the familiar matrix form
\[ \begin{aligned} \ddot D_1(\ddot\upsilon_T-\overline\upsilon_{T*}) &= \left[\ddot D_1^{-1}\sum\ddot r_{t-1}\ddot r_{t-1}’\ddot D_1^{-1}\right]^{-1} \left[\ddot D_1^{-1}\sum\ddot r_{t-1}u_{t-1}\right]\\ &\Rightarrow\mathcal L:=\mathcal M_{11}^{-1}\mathcal S. \end{aligned} \]The Brownian paths summarize how the integrated regressors vary over the entire sample. Their randomness remains in the moment matrix as well as in the score. Both objects therefore enter the coefficient distribution jointly; replacing the random moment matrix by its expectation would change the limit.
The dimension calculation in the OLS formula is the same on both sides. The normalized cross-product matrix and \(\mathcal M_{11}^{-1}\) are \((2k+2)\times(2k+2)\); the normalized score and its limit \(\mathcal S\) are \((2k+2)\times1\). Their product is therefore the \((2k+2)\times1\) coefficient-error limit \(\mathcal L\). Its first block \(\mathcal L_1\) has \(2k\) entries; \(\mathcal L_2\) and \(\mathcal L_3\) are scalars.
Lemma 3(i) partitions \(\mathcal L=(\mathcal L_1′,\mathcal L_2,\mathcal L_3)’\). Reading its normalization component by component gives
The second line of Equation (F.3) measures the error relative to \(\vartheta_{T*}\), the coefficient required by the particular detrended regressors in that sample. To measure the difference relative to the fixed limit \(\vartheta_*\), add and subtract that sample-dependent target:
\[ \begin{aligned} \ddot\vartheta_T-\vartheta_* &=(\ddot\vartheta_T-\vartheta_{T*})+(\vartheta_{T*}-\vartheta_*)\\ &=(\ddot\vartheta_T-\vartheta_{T*}) +\beta_*^{\prime}(\ddot\mu_T-\mu_*). \end{aligned} \]The first term has order \(T^{-3/2}\); the second generally has the larger order \(T^{-1/2}\). The precise estimation of a sample-dependent trend coefficient can therefore coexist with a less precise comparison to its fixed limiting value. Both differences must be identified before a convergence rate is interpreted.
For the slope, the statement \(T(\ddot\beta_T-\beta_*)\Rightarrow\mathcal L_1\) means that multiplication by \(T\) preserves a nontrivial limiting fluctuation. This implies that the unscaled error is of order \(T^{-1}\). The intercept requires only \(\sqrt T\), corresponding to an error of order \(T^{-1/2}\). These rates are read from the normalization of the coefficient errors after the limiting matrix and score have been established.
The arrow \(\Rightarrow\) has the distributional meaning defined in Section 7: the probabilities attached to the scaled error approach those specified by its limiting random vector. For example, for one slope component, the probability that \(T(\ddot\beta_{j,T}-\beta_{j,*})\) falls in a given interval approaches the corresponding probability under \(\mathcal L_{1,j}\), provided the interval endpoints carry no probability mass in that limiting distribution. The unscaled error shrinks; the multiplier keeps its uncertainty visible.
Multiplication by \(T\) magnifies the shrinking slope error. Without that multiplication, the error converges to zero in probability; with it, the uncertainty remains visible in a stable limiting distribution. Brownian paths describe the randomness retained from the integrated regressors over the sample. This is why the transformed moment matrix and the coefficient-error limit remain random even though the original normalized matrix in Appendix C converged to a fixed matrix.
F.3. Read the stochastic-integral vector
The cross-product matrix adds squared regressors. The score instead adds each regressor multiplied by the error \(u_{t-1}\). Its limit therefore involves increments of the cumulative-error path. To read this distinction, define
\[ U_T(r):=\frac1{\sqrt T}\sum_{j=1}^{\lfloor Tr\rfloor}u_{j-1}, \qquad U_T(t/T)-U_T((t-1)/T)=\frac{u_{t-1}}{\sqrt T}. \]This shifted cumulative-error process has the same Brownian limit as the cumulative process dated \(u_t\) under the maintained assumptions. In the trend score, substituting \(r_{t,T}=q_t/T\) gives the exact weighted-increment sum
\[ \frac1{T^{3/2}}\sum_{t=1}^{T}q_tu_{t-1} =\sum_{t=1}^{T}r_{t,T}\bigl[U_T(t/T)-U_T((t-1)/T)\bigr] \Rightarrow\int_0^1r\,d\mathcal B_u(r). \]The integrator is \(d\mathcal B_u\) because the quantities being added are weighted changes in a random cumulative process. For the constant score the weight is one; summing all increments leaves the endpoint \(U_T(1)\). For a detrended partial sum the weight is \(\ddot m_{t-1}/\sqrt T\), which is itself random:
\[ \frac1T\sum_{t=1}^{T}\ddot m_{t-1}u_{t-1} =\sum_{t=1}^{T}\frac{\ddot m_{t-1}}{\sqrt T} \bigl[U_T(t/T)-U_T((t-1)/T)\bigr]. \]The dependence between that weight and the errors must be retained. Lemma 3 supplies the resulting limit, including the dependence correction displayed below.
Let \(\mathcal B_u(r)\) denote the Brownian limit of the normalized cumulative cointegration errors, \(T^{-1/2}\sum_{t=1}^{\lfloor Tr\rfloor}u_t\). It is considered jointly with \(\mathcal B_s(r)\); the two processes need not be independent. With every integral below taken over \([0,1]\), the three score limits are
\[ \begin{aligned} \mathcal S_1&=\int\mathcal B_s\,d\mathcal B_u -3\left(\int r\mathcal B_s\,dr\right)\left(\int r\,d\mathcal B_u\right) +\varrho_{m*},\\ \mathcal S_2&=\int r\,d\mathcal B_u,\qquad \mathcal S_3=\int d\mathcal B_u. \end{aligned} \]An integral with \(dr\) accumulates a random path over time. An integral with \(d\mathcal B_u\) accumulates weighted Brownian increments. In particular, \(\int d\mathcal B_u=\mathcal B_u(1)\). The first component combines the centered partial sums with the cointegration error; its subtraction accounts for the fitted trend.
In this score vector, \(\mathcal B_u\) is scalar. Hence \(\mathcal S_1\) is \(2k\times1\), whereas \(\mathcal S_2\) and \(\mathcal S_3\) are scalars. The correction \(\varrho_{m*}\) below is \(2k\times1\), since each product \(s_ju_t\) multiplies a \(2k\times1\) increment vector by a scalar error. Together the three blocks form \(\mathcal S=(\mathcal S_1′,\mathcal S_2,\mathcal S_3)’\), with \(2k+2\) entries.
Cho’s Lemma 3 writes the dependence correction as
\[ \varrho_{m*}:=\lim_{T\to\infty}\frac1T \sum_{t=1}^{T}\sum_{j=1}^{t-1}E[s_ju_t], \qquad s_j=\Delta\ddot x_j-\mu_*. \]It records dependence between past centered increments and the cointegration error. This contributes to the asymptotic bias component of the normalized limit. Consistency remains compatible with that contribution because the unscaled slope error shrinks at rate \(T^{-1}\). The primitive limit is generally nonstandard; mixed normality of the final one-step estimator is established later.
F.4. Recover the original trend coefficient
Combining the definition of \(\vartheta_{T*}\) with Proposition 2 gives an exact cancellation:
\[ \begin{aligned} \widetilde\zeta_T-\zeta_* &=\ddot\vartheta_T-\ddot\mu_T^{\prime}\ddot\beta_T-\zeta_*\\ &=(\ddot\vartheta_T-\vartheta_{T*}) +(\zeta_*+\ddot\mu_T^{\prime}\beta_*) -\ddot\mu_T^{\prime}\ddot\beta_T-\zeta_*\\ &=(\ddot\vartheta_T-\vartheta_{T*}) -\ddot\mu_T^{\prime}(\ddot\beta_T-\beta_*). \end{aligned} \]After multiplication by \(T\), the first term is \(O_p(T^{-1/2})\), while the second retains the slope-estimation limit. Therefore
Equation (F.4) explains Lemma 3(ii): the original slope and explicit-trend estimates share \(\mathcal L_1\), while the intercept retains \(\mathcal L_3\). Restoring the original partial sums reallocates their estimated trend contribution to the explicit trend coefficient.
The replacement normalizer \(\ddot D_\dagger\) is again \((2k+2)\times(2k+2)\); only the trend’s multiplier changes. The limit \(-\mu_*’\mathcal L_1\) is scalar because it multiplies a \(1\times2k\) row by a \(2k\times1\) column.
Main takeaway. The transformed moment matrix retains the random variation of the detrended partial sums and has a nonsingular limit. This gives the auxiliary coefficient distributions; the trend coefficient must be centered at the target implied by the fitted drift.
Source: Cho, Lemmas 2–3. The Gram matrix above is written directly from the stated detrended Brownian path. Return to the main text: End of Section 5: the transformed matrix · End of Section 7: probability limits · Contents.
Appendix G. Lemma 4: account for the estimated cointegration residual
Return to the main text: End of Section 7: generated regressors and Lemma 4 · Contents.
G.1. The generated regressor carries first-stage estimation error
The dynamic regression uses \(\ddot u_{t-1}\), whereas the population error-correction equation contains \(u_{t-1}\). Their difference is determined exactly by the auxiliary long-run estimates. Put
\[ b_T=\ddot\beta_T-\beta_*,\qquad v_T=\ddot\vartheta_T-\vartheta_{T*},\qquad n_T=\ddot\nu_T-\nu_*. \]Both residual definitions use the same transformed regressors. Write them on separate lines before subtracting:
\[ \begin{aligned} u_{t-1}&=y_{t-1}-\beta_*^{\prime}\ddot m_{t-1}-\vartheta_{T*}q_t-\nu_*,\\ \ddot u_{t-1}&=y_{t-1}-\ddot\beta_T^{\prime}\ddot m_{t-1}-\ddot\vartheta_Tq_t-\ddot\nu_T. \end{aligned} \]The \(y_{t-1}\) terms cancel. Subtracting the second line from the first changes the signs of the second line’s subtracted fitted values:
\[ \begin{aligned} u_{t-1}-\ddot u_{t-1} &=(\ddot\beta_T-\beta_*)^{\prime}\ddot m_{t-1} +(\ddot\vartheta_T-\vartheta_{T*})q_t +(\ddot\nu_T-\nu_*)\\ &=b_T^{\prime}\ddot m_{t-1}+v_Tq_t+n_T. \end{aligned} \]Adding the fitted residual back gives
\[ u_{t-1}=\ddot u_{t-1}+b_T’\ddot m_{t-1}+v_Tq_t+n_T. \]Consequently, the equation \(\Delta y_t=\rho_*u_{t-1}+\gamma_*+\kappa_*’z_{2t}+e_t\) can be expressed in the generated regressors as
\[ \Delta y_t=\rho_*\ddot u_{t-1} +\rho_*b_T’\ddot m_{t-1}+\rho_*v_Tq_t +\gamma_*+\rho_*n_T+\kappa_*’z_{2t}+e_t. \]The two target vectors in Equation (G.1) keep track of these changes. The retained levels and trend have sample-dependent coefficients. Their fixed population targets are zero, but their coefficients in this exact reparameterization reflect first-stage estimation error. This is why their contribution must be followed through the asymptotic calculation.
For a concrete scalar illustration, suppose the estimated long-run slope exceeds its true value by \(0.02\), and \(\rho_*=-0.5\). The fitted cointegration residual subtracts too much of that regressor. Restoring the true dynamic equation requires a coefficient \(\rho_*b_T=-0.01\) on its retained detrended level. This coefficient compensates exactly for the error already embedded in the generated residual.
Collecting these coefficients gives the sample-dependent target \(\tau_{T*}\) and its fixed population counterpart \(\tau_*\):
Dimensionally, \(b_T\) is \(2k\times1\), while \(v_T\) and \(n_T\) are scalars. The product \(b_T’\ddot m_{t-1}\) is therefore scalar. Both target vectors displayed above have block lengths \(1,2k,1,1,s\), giving \(K=2k+s+3\) entries. The notation \(0_{2k}\) denotes a \(2k\times1\) column of zeros.
G.2. Part (i): center at the coefficient vector of the generated-regressor equation
Stack the transformed regressors as \(\ddot z_t=(\ddot u_{t-1},\ddot m_{t-1}’,q_t,1,z_{2t}’)’\). Their column normalizer is
\[ \ddot D=\operatorname{diag}(T^{1/2},TI_{2k},T^{3/2},T^{1/2},T^{1/2}I_s). \]The first block corresponds to the cointegration residual, the next to integrated fluctuations, and the third to the deterministic trend. The constant and short-run regressors complete the stationary blocks.
The regressor vector \(\ddot z_t\), coefficient vector \(\ddot\tau_T\), and target vector \(\tau_{T*}\) are each \(K\times1\). Stacking the rows \(\ddot z_t’\) over all dates gives a \(T\times K\) regression matrix. Both \(\ddot D\) and the summed cross-product \(\sum_t\ddot z_t\ddot z_t’\) are \(K\times K\). The score \(\sum_t\ddot z_te_t\) is \(K\times1\). Thus the inverse matrix times the score has the required \(K\times1\) size. Normalizing the exact OLS identity gives
Full sample rank gives the finite-sample inverse in Equation (G.2). Lemma 2 establishes that its normalized matrix has a limit \(\mathcal M\) that is nonsingular with probability one, allowing the inverse to be taken in the limiting argument. The random vector \(\mathcal D\), distinct from the normalization matrix \(\ddot D\), has five blocks:
\[ \mathcal D=(\mathcal D_1,\mathcal D_2′,\mathcal D_3,\mathcal D_4,\mathcal D_5′)’, \quad \mathcal J= \begin{pmatrix} \int d\mathcal B_{ue}\\ \int\overline{\mathcal B}_s\,d\mathcal B_e\\ \int r\,d\mathcal B_e\\ \int d\mathcal B_e\\ \int d\mathcal B_{ze} \end{pmatrix}. \]Each score block pairs its regressor with the innovation \(e_t\). Here \(\mathcal B_e(r)\) is the Brownian limit of \(T^{-1/2}\sum_{t=1}^{\lfloor Tr\rfloor}e_t\). The processes \(\mathcal B_{ue}\) and \(\mathcal B_{ze}\) arise from normalized cumulative stationary products \(u_{t-1}e_t\) and \(z_{2t}e_t\). The subscript \(ue\) labels a process; it does not mean a product of two Brownian motions.
The matrix \(\mathcal M\) is \(K\times K\), while \(\mathcal J\) and \(\mathcal D=\mathcal M^{-1}\mathcal J\) are \(K\times1\). Their five block sizes are \(1,2k,1,1,s\). Accordingly, \(\mathcal D_1,\mathcal D_3,\mathcal D_4\) are scalars, \(\mathcal D_2\) is \(2k\times1\), and \(\mathcal D_5\) is \(s\times1\). The processes \(\mathcal B_e\) and \(\mathcal B_{ue}\) are scalar; \(\mathcal B_{ze}\) is \(s\times1\).
The second entry is the regression-limit expression
\[ \int\overline{\mathcal B}_s\,d\mathcal B_e :=\int\mathcal B_s\,d\mathcal B_e -3\left(\int r\mathcal B_s\,dr\right)\left(\int r\,d\mathcal B_e\right). \]This expanded definition retains the effect of estimating the trend from the entire sample. Lemma 4(i) establishes mixed normality: conditional on the relevant limiting regressor information, the distribution is Gaussian, with a covariance matrix that can depend on that information.
G.3. Part (ii): change from the random target to the fixed target
The change of centering follows an elementary addition:
The second term in Equation (G.3) is found by subtracting the two target vectors in Equation (G.1) entry by entry. The adjustment and short-run blocks cancel, while the other blocks retain the first-stage errors:
\[ \tau_{T*}-\tau_* =\rho_*\begin{pmatrix}0\\b_T\\v_T\\n_T\\0_s\end{pmatrix}. \]Multiplying by the diagonal matrix \(\ddot D\) scales each block separately:
\[ \begin{aligned} \ddot D(\tau_{T*}-\tau_*) &=\rho_*\begin{pmatrix} \sqrt T\,0\\TI_{2k}b_T\\T^{3/2}v_T\\\sqrt T n_T\\\sqrt T I_s0_s \end{pmatrix} =\rho_*\begin{pmatrix}0\\Tb_T\\T^{3/2}v_T\\\sqrt T n_T\\0_s\end{pmatrix}\\ &\Rightarrow\rho_*\begin{pmatrix}0\\\mathcal L_1\\\mathcal L_2\\\mathcal L_3\\0_s\end{pmatrix} =\rho_*[0,\mathcal L^{\prime},0_s^{\prime}]^{\prime}. \end{aligned} \]The middle three limits are exactly the three limits in Lemma 3. Adding this vector to the centered dynamic limit, using their joint convergence, gives
The limits \(\mathcal D\) and \(\mathcal L\) in Equation (G.4) originate from the same observations and must be treated jointly. Their dependence matters for the cancellations in the final OLS estimator. Centering at \(\tau_*\) generally adds the nonstandard first-stage component, so the mixed-normal conclusion of part (i) cannot simply be transferred to this differently centered vector.
The added vector \([0,\mathcal L’,0′]’\) has the same length as \(\mathcal D\): its first zero is scalar, its middle block contains the \(2k+2\) entries of \(\mathcal L\), and its last zero block has \(s\) entries. The total is \(1+(2k+2)+s=K\). It inserts the first-stage slope, trend, and intercept limits in their corresponding dynamic-regression positions.
The distinction can also be read coefficient by coefficient. The adjustment coefficient and short-run coefficients have the same targets in \(\tau_{T*}\) and \(\tau_*\). Their scaled limits remain \(\mathcal D_1\) and \(\mathcal D_5\). The intervening blocks acquire \(\rho_*\mathcal L_1\), \(\rho_*\mathcal L_2\), and \(\rho_*\mathcal L_3\). These additions arise because the estimated residual carries slope, trend, and intercept estimation error simultaneously.
Returning to the scalar illustration, suppose the estimated coefficient on the retained level is \(-0.006\). Its deviation from the random target \(-0.01\) is \(0.004\); its deviation from the fixed target zero is \(-0.006\). The centering identity reads \(-0.006=0.004+(-0.01)\). The first-stage adjustment is added with its own sign. The same accounting applies to every affected coordinate of the normalized vector.
Note on the printed sign in Lemma 4(ii). The displayed lemma prints a minus sign before \(\rho_*[0,\mathcal L’,0′]’\). This appears inconsistent with the definitions of \(\tau_{T*}\) and \(\mathcal L\): the exact centering identity above requires a plus sign. The formula used here follows that algebra; the negative value of \(\rho_*\) does not change the addition rule.
Main takeaway. The estimated cointegration residual carries first-stage estimation error into the dynamic equation. Tracking that error in the coefficient target gives the correct joint limit and prepares the cancellations used to reconstruct OLS.
Source: Cho, Section 4.2 and Lemma 4, with the sign discrepancy identified explicitly above. Return to the main text: End of Section 7: generated regressors and Lemma 4 · Contents.
Appendix H. From the primitive limits to the original OLS estimator
Return to the main text: End of Section 8: coefficient convergence rates · End of Section 9: implications for applied work · Contents.
H.1. The exact product expansion
Proposition 4 gives \(\widehat{\boldsymbol\alpha}_T=\mathbf T_T\ddot\tau_T\). Its population counterpart is \(\boldsymbol\alpha_* =\mathbf T\tau_*\), where \(\mathbf T\) is the probability limit of \(\mathbf T_T\). Write each estimated object as its target plus its estimation error, then distribute the product:
\[ \begin{aligned} \mathbf T_T\ddot\tau_T &=\bigl[\mathbf T+(\mathbf T_T-\mathbf T)\bigr] \bigl[\tau_*+(\ddot\tau_T-\tau_*)\bigr]\\ &=\mathbf T\tau_*+\mathbf T(\ddot\tau_T-\tau_*)\\ &\quad+(\mathbf T_T-\mathbf T)\tau_* +(\mathbf T_T-\mathbf T)(\ddot\tau_T-\tau_*). \end{aligned} \]Every product keeps the matrix on the left of the vector. Subtracting \(\mathbf T\tau_*\) removes the first term and leaves
In Equation (H.1), the first term is error in the transformation, the second is error in the primitive estimator, and the third is their product. The matrices and coefficient vectors are estimated jointly. Their orders can nevertheless be multiplied without assuming independence: globally, the last term is \(O_p(T^{-1})\), so multiplying it by \(\sqrt T\) makes it vanish in probability.
With \(K=2k+s+3\) coefficients in the full model, both transformation matrices \(\mathbf T_T\) and \(\mathbf T\) are \(K\times K\); every coefficient vector in this identity is \(K\times1\). Each term therefore multiplies a \(K\times K\) matrix by a \(K\times1\) vector, yielding a \(K\times1\) vector that can be added to the other two.
H.2. Identify the leading error coefficient by coefficient
Put \(a_T=\ddot\rho_T-\rho_*\), retaining \(b_T\) and \(n_T\) from Appendix G. Thus \(\ddot\rho_T=\rho_*+a_T\) and \(\ddot\beta_T=\beta_*+b_T\). Substitute these expressions into \(\widehat\theta_T=\ddot\eta_T-\ddot\rho_T\ddot\beta_T\), and use \(\theta_*=-\rho_*\beta_*\):
The two terms containing \(\rho_*\beta_*\) in Equation (H.2) cancel. The bracket \(\ddot\eta_T-\rho_*b_T\) is the dynamic-coefficient error centered at its sample-dependent target, as in Lemma 4(i).
The three terms have orders \(T^{-1/2}\), \(T^{-1}\), and \(T^{-3/2}\) when the corresponding component of \(\beta_*\) is nonzero. To show what survives, write \(h_T:=\ddot\eta_T-\rho_*b_T\) and multiply every term by \(\sqrt T\):
\[ \sqrt T(\widehat\theta_T-\theta_*) =-\beta_*\sqrt T a_T +\frac{Th_T}{\sqrt T} -\frac{(\sqrt T a_T)(Tb_T)}{T}. \]Lemma 4 gives finite distributional limits for \(\sqrt T a_T\) and \(Th_T\); Lemma 3 does so for \(Tb_T\). Dividing the latter two contributions by the displayed growing denominators makes them vanish in probability. The surviving expression is
\[ \sqrt T(\widehat\theta_T-\theta_*) =-\beta_*\sqrt T a_T+o_p(1) \Rightarrow-\beta_*\mathcal D_1. \]The dominant uncertainty comes from estimating the adjustment coefficient. The long-run slope estimate converges faster, and its remaining contribution is combined with \(\ddot\eta_T\) in the bracket governed by Lemma 4(i). Retaining the product term above keeps the expansion exact before the asymptotic simplification.
The slope expansion is a vector identity: \(\widehat\theta_T\), \(\theta_*\), \(\beta_*\), \(\ddot\eta_T\), and \(b_T\) are all \(2k\times1\). The adjustment error \(a_T\) and limit \(\mathcal D_1\) are scalars, so multiplication by them preserves this vector size. The intercept and trend expansions that follow are scalar identities; the short-run coefficient identity is \(s\times1\).
The powers can be compared numerically without assigning values to the stochastic coefficients. At \(T=100\), the magnitudes \(T^{-1/2}\), \(T^{-1}\), and \(T^{-3/2}\) are \(0.1\), \(0.01\), and \(0.001\). At \(T=400\), they are \(0.05\), \(0.0025\), and \(0.000125\). The second and third scales shrink more quickly. Multiplying by \(\sqrt T\) preserves the first scale while the others approach zero.
The intercept follows the same product calculation. Here \(\widehat\alpha_T=\ddot\gamma_T-\ddot\rho_T\ddot\nu_T\), \(\alpha_*=\gamma_*-\rho_*\nu_*\), and \(\ddot\nu_T=\nu_*+n_T\). Therefore
\[ \begin{aligned} \widehat\alpha_T-\alpha_* &=(\ddot\gamma_T-\gamma_*)-(\rho_*+a_T)(\nu_*+n_T)+\rho_*\nu_*\\ &=(\ddot\gamma_T-\gamma_*)-\rho_*n_T-\nu_*a_T-a_Tn_T\\ &=-\nu_*a_T+[(\ddot\gamma_T-\gamma_*)-\rho_*n_T]-a_Tn_T. \end{aligned} \]The bracket has order \(T^{-1/2}\) and scaled limit \(\mathcal D_4\). It subtracts the first-stage intercept error embedded in the generated-regressor coefficient. The final product is \(O_p(T^{-1})\).
For the trend, carry both corrections through the calculation. Define its centered dynamic error as \(g_T:=\ddot\varsigma_T-\rho_*v_T\), and retain \(h_T=\ddot\eta_T-\rho_*b_T\). Lemma 4 gives \(g_T=O_p(T^{-3/2})\) and \(h_T=O_p(T^{-1})\). The auxiliary trend identity in Appendix F gives
\[ \widetilde\zeta_T-\zeta_*=v_T-\ddot\mu_T’b_T. \]Use \(\xi_*=-\rho_*\zeta_*\), expand the reconstructed trend coefficient, and substitute this identity:
\[ \begin{aligned} \widehat\xi_T-\xi_* &=\ddot\varsigma_T-\ddot\mu_T’\ddot\eta_T -(\rho_*+a_T)\widetilde\zeta_T+\rho_*\zeta_*\\ &=\ddot\varsigma_T-\ddot\mu_T’\ddot\eta_T -\rho_*(\widetilde\zeta_T-\zeta_*)-a_T\widetilde\zeta_T\\ &=(\ddot\varsigma_T-\rho_*v_T) -\ddot\mu_T'(\ddot\eta_T-\rho_*b_T)-a_T\widetilde\zeta_T\\ &=-\zeta_*a_T+g_T-\ddot\mu_T’h_T -a_T(\widetilde\zeta_T-\zeta_*). \end{aligned} \]For \(\zeta_*\ne0\), these four terms have orders \(O_p(T^{-1/2})\), \(O_p(T^{-3/2})\), \(O_p(T^{-1})\), and \(O_p(T^{-3/2})\), respectively. The third term uses \(\ddot\mu_T=O_p(1)\); the last multiplies an \(O_p(T^{-1/2})\) adjustment error by an \(O_p(T^{-1})\) auxiliary trend error. Multiplication by \(\sqrt T\) therefore leaves the limit \(-\zeta_*\mathcal D_1\).
If \(\zeta_*=0\), the first term is exactly zero. Multiplication by \(T\) then retains the third term, giving \(T(\widehat\xi_T-\xi_*)\Rightarrow-\mu_*’\mathcal D_2\); the other two terms vanish at that scale. This calculation explains the finer trend rate reported in the table below. The short-run coefficients are copied unchanged by the reconstruction:
\[ \widehat\kappa_T-\kappa_*=\ddot\kappa_T-\kappa_*. \]H.3. Collect the surviving terms
The first column of \(\mathbf T\) collects the coefficients multiplying the adjustment error:
\[ c_*=(1,-\beta_*’,-\zeta_*,-\nu_*,0′)’, \quad d_T= \begin{pmatrix} 0\\0_{2k}\\0\\ (\ddot\gamma_T-\gamma_*)-\rho_*(\ddot\nu_T-\nu_*)\\ \ddot\kappa_T-\kappa_* \end{pmatrix}. \]Stacking the component expansions gives Cho’s equation (16). When every relevant component of \(\beta_*^+\), \(\beta_*^-\), and \(\zeta_*\) is nonzero, this yields the generic normalization in Theorem 1:
In Equation (H.3), the first-stage terms have canceled from the final limit. The remaining vector is determined by \(\mathcal D\) and is mixed normal. Several coefficients share the same dominant random component \(\mathcal D_1\), which explains the dependence of their limiting estimation errors.
Both \(c_*\) and \(d_T\) have \(K\) entries, partitioned as \(1,2k,1,1,s\). The trailing zero in \(c_*\) has \(s\) entries. In the displayed limit, \(c_*\mathcal D_1\) is a \(K\times1\) vector and the added column also has \(K\) entries: one zero, \(2k\) zeros, one zero, the scalar \(\mathcal D_4\), and the \(s\times1\) block \(\mathcal D_5\). This verifies that each original coefficient receives exactly one corresponding limit entry.
The common component has a direct interpretation. Holding the long-run slopes fixed, changing the estimated adjustment speed changes every equation-level coefficient through \(\theta_*=-\rho_*\beta_*\). Their leading estimation errors therefore move together, with loadings given by the long-run coefficients. The smaller terms still distinguish the separate estimates within an observed sample and become decisive when a leading loading vanishes.
H.4. A zero leading coefficient changes the informative rate
If a component of \(\beta_*\) equals zero, its contribution \(-\beta_*a_T\) is exactly zero. For the theorem’s case in which the entire positive block is zero, the preceding expansion becomes
\[ \begin{aligned} \widehat\theta_T^+-\theta_*^+ &=0+h_T^+-a_Tb_T^+,\\ T(\widehat\theta_T^+-\theta_*^+) &=Th_T^+-\frac{(\sqrt T a_T)(Tb_T^+)}{\sqrt T} \Rightarrow\mathcal D_2^+. \end{aligned} \]The \(T^{-1}\) centered error now supplies the leading random variation. Multiplication by \(\sqrt T\) would give a zero limit; multiplication by \(T\) keeps that variation visible. A nondegenerate limit means a distribution retaining random variation, rather than collapsing to a single value. The trend coefficient requires the same finer multiplier when \(\zeta_*=0\). The four cases stated in Theorem 1 are summarized below; parameters not set to zero satisfy the theorem’s stated nonzero conditions.
| Population-parameter case | Positive level error \(\widehat\theta_T^+-\theta_*^+\) | Negative level error \(\widehat\theta_T^–\theta_*^-\) | Trend error \(\widehat\xi_T-\xi_*\) | Adjustment, intercept, and short-run errors |
|---|---|---|---|---|
| All relevant \(\beta_*^\pm\) and \(\zeta_*\) nonzero | \(\sqrt T\) | \(\sqrt T\) | \(\sqrt T\) | \(\sqrt T\) |
| \(\beta_*^+=0\) | \(T\) | \(\sqrt T\) | \(\sqrt T\) | \(\sqrt T\) |
| \(\beta_*^-=0\) | \(\sqrt T\) | \(T\) | \(\sqrt T\) | \(\sqrt T\) |
| \(\zeta_*=0\) | \(\sqrt T\) | \(\sqrt T\) | \(T\) | \(\sqrt T\) |
Table notes. Choose a row according to the population-parameter case described by the theorem, then apply the cell’s multiplier to the estimation error named in that column. For example, the positive-level cell of the second row means \(T(\widehat\theta_T^+-\theta_*^+)\), whose limit is derived immediately above. The final column applies separately to \(\widehat\rho_T-\rho_*\), \(\widehat\alpha_T-\alpha_*\), and \(\widehat\kappa_T-\kappa_*\).
The connection between a multiplier and an error scale follows from cancellation: \(T\times T^{-1}=1\), while \(\sqrt T\times T^{-1/2}=1\). Thus the larger multiplier \(T\) reveals a smaller estimation error. These are statements about the theorem’s true parameter configurations. A small estimated coefficient does not establish that its true value is exactly zero and is not a rule for switching between the table’s normalizations.
For \(k>1\), a zero \(\beta_*^+\) or \(\beta_*^-\) in these listed cases means the corresponding whole block is zero. The faster limits are \(\mathcal D_2^+\), \(\mathcal D_2^-\), and \(-\mu_*’\mathcal D_2\), respectively. The \(\sqrt T\) expansion still holds in those directions, but yields zero; the finer normalization reveals their remaining variation.
The positive and negative slope blocks \(\widehat\theta_T^+\) and \(\widehat\theta_T^-\) are each \(k\times1\). Their limit blocks \(\mathcal D_2^+\) and \(\mathcal D_2^-\) also have \(k\) entries and together form the \(2k\times1\) vector \(\mathcal D_2\). The trend limit \(-\mu_*’\mathcal D_2\) is scalar. The long-run ratio below is a \(2k\times1\) vector divided by a scalar adjustment coefficient, so its limit has \(2k\) entries.
The mechanism can be represented by an error of the form \(a/\sqrt T+b/T\). When the leading coefficient \(a\) is nonzero, the first term dominates, and the \(\sqrt T\) multiplier preserves it. If \(a=0\), the error becomes \(b/T\), and multiplication by \(T\) is needed to preserve a nonzero limit. For the estimator, the error terms also contain random factors, while a zero population loading removes the first contribution.
The zero-parameter cases therefore describe changes in which source of estimation uncertainty is dominant. They do not require changing the computational OLS procedure. The coefficient identity remains the same; its expansion reveals that a term previously of smaller order now provides the leading information. This is why the theorem states separate parameter cases instead of assigning one informative normalization to every possible coefficient configuration.
Comparisons with two-step NARDL should keep the target parameter fixed. A two-step estimator of \(\beta_*\) converges at rate \(T\), while its estimator of \(\rho_*\) has rate \(\sqrt T\). Their product estimates \(\theta_*=-\rho_*\beta_*\) at the generic \(\sqrt T\) rate, matching the corresponding one-step coefficient.
A further algebraic deduction concerns the one-step long-run ratio. For fixed \(\rho_*\ne0\), define \(\widehat\beta_T^{\mathrm{ratio}}=-\widehat\theta_T/\widehat\rho_T\). Begin with the coefficient reconstruction before replacing any errors by their limits:
\[ \begin{aligned} \widehat\beta_T^{\mathrm{ratio}} &=-\frac{\ddot\eta_T-\ddot\rho_T\ddot\beta_T}{\ddot\rho_T} =\ddot\beta_T-\frac{\ddot\eta_T}{\ddot\rho_T},\\ \widehat\beta_T^{\mathrm{ratio}}-\beta_* &=b_T-\frac{\ddot\eta_T}{\ddot\rho_T} =\frac{\ddot\rho_Tb_T-\ddot\eta_T}{\ddot\rho_T}\\ &=\frac{(\rho_*+a_T)b_T-\ddot\eta_T}{\ddot\rho_T} =\frac{a_Tb_T-[\ddot\eta_T-\rho_*b_T]}{\ddot\rho_T} =\frac{a_Tb_T-h_T}{\ddot\rho_T}. \end{aligned} \]The leading adjustment-error term that appeared in the level coefficient has canceled. Multiplying the remaining difference by \(T\) gives
In Equation (H.4), the first numerator term vanishes in probability, \(Th_T\) converges in distribution to \(\mathcal D_2\), and the denominator converges in probability to the fixed nonzero \(\rho_*\). These three facts explain the final limit.
This deduction from the identities and lemmas distinguishes the superconsistent long-run ratio from the equation coefficient \(\widehat\theta_T\). The common adjustment error cancels in the ratio. Throughout these results, the maintained cointegrating framework has \(\rho_*<0\).
Main takeaway. The exact product expansion identifies which estimation error governs each original coefficient. It explains the common adjustment-error component, the faster rates when that component vanishes, and its cancellation in the long-run ratios.
Source: Cho, Section 4.3, equations (15)–(17), and Theorem 1. The ratio result is an explicit deduction from the preceding identities. Return to the main text: End of Section 8: coefficient convergence rates · End of Section 9: implications for applied work · Contents.