The PDL_REGRESSION command estimates a Polynomial Distributed Lag model. It measures how changes in one time series affect another over multiple lag periods, while forcing the lag weights to follow a smooth polynomial pattern.
PDL_REGRESSION(LIST(CPIAUCSL,M2SL))This example estimates the lagged relationship between consumer prices and the money supply.
PDL_REGRESSION(seriesList)The input should be a list containing two time series. The first series is the dependent variable, and the second series is the explanatory variable.
The command returns a full PDL analysis, including the fitted model, actual versus fitted values, period lag weights, cumulative lag weights, and a text explanation of the model.
A Polynomial Distributed Lag model is useful when the effect of one variable is not immediate. For example, changes in money supply, interest rates, energy prices, or employment conditions may influence inflation, output, or market prices over several future periods.
Rather than estimating a separate coefficient for every lag, the PDL model constrains the lag coefficients to follow a polynomial curve. This reduces overfitting and makes the lag structure easier to interpret.
The lag weights show how much influence the explanatory variable has at each lag. Positive weights indicate a positive delayed effect, while negative weights indicate a negative delayed effect.
The cumulative weights show the total effect over time. This is often the most useful chart for judging whether the explanatory variable has a lasting impact.
The model uses a maximum lag, polynomial degree, and decay factor to estimate a smooth lag pattern. The decay factor allows the influence of older lags to diminish over time.