Approximately Normal Tests for Equal Predictive Accuracy in Nested Models
read more
Citations
Policy Uncertainty and Corporate Investment
Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy
A Very Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix
Investor sentiment aligned: : A powerful predictor of stock returns
Forecasting the Equity Risk Premium: The Role of Technical Indicators
References
A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix
Comparing Predictive Accuracy
A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelationconsistent Covariance Matrix
Comparing Predictive Accuracy
Dividend yields and expected stock returns
Related Papers (5)
Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy
A Comprehensive Look at the Empirical Performance of Equity Premium Prediction
Frequently Asked Questions (11)
Q2. What is the commonly used statistic for comparisons of predictions from nested models?
Perhaps the most commonly used statistic for comparisons of predictions from nested models is mean squared prediction error (MSPE).1
Q3. What is the common reason for the oversizing of Clark and McCracken?
The occasional oversizing Clark and McCracken (2001, 2005a) find arises when data-determined lag selection yields significantly misspecified null forecasting models.
Q4. What is the effect of the adjusted MSPE test on the null model?
The results for their adjusted MSPE test highlight the potential for noise associated with theadditional parameters of the alternative model to create an upward shift in the model’s MSPE large enough that the null model has a lower MSPE even when the alternative model is true.
Q5. What is the conditional variance of the predictand yt+1?
In panel B, the predictand yt+1 has conditional heteroskedasticity of the form given in equation (6.3), in which the conditional variance at t is a function of z2t-1.
Q6. What is the argument about the standard error in nested models?
The authors are about to argue that in nested models, conventional standard errors yield an asymptotic normal approximation that is accurate for practical purposes.
Q7. What is the mean value of the squared difference in fitted values?
Across simulations, the implied mean value of the squared difference in fitted values P-13 tT=R( ^y1t+1^y2t+1) 2 is 0.25 (=0.01-(-0.24)).
Q8. What is the chi-squared test statistic associated with (5.3)?
In the notation of (3.1) and (3.2), the null and sample moment used to test the null are:Ee1tZtN=0, (5.2) P-13 tT=R ^e1t+1Zt+1N. (CCS) (5.3)The chi-squared test statistic associated with (5.3) was adjusted for uncertainty due to estimation of regression parameters as described in Chao et al. (2001).
Q9. What is the approximation of R fixed?
The approximation that the authors have just discussed, which holds R fixed as P goes to infinity, therebyimplying R/P goes to 0, may not be obviously appealing.
Q10. What is the way to compare the bootstrap with MSPE?
The authors find that on balance, the bootstrap performs distinctly better than MSPE-adjusted for relatively small samples sizes, comparably for medium or larger sample sizes; overall, MSPE-adjusted performs a little better than CCS, a lot better than MSPE-normal.
Q11. What is the way to compare a parsimonious model to a larger alternative?
The larger alternative model might be a bivariate or multivariate vector autoregression (VAR) that includes lags of some variables in addition to lags of the variable to be predicted.