Saturday, May 28, 2005

Robert Waldmann’s Stalled Projects Without Co-Authors which have not been presented at informal talks the next of which might happen in Room C Thursday June 9 at 15:00.
If anyone wants to hear about any of these topics (except 4 because I am not going to talk about inequality data again that soon) vote in comments or by e-mail.

1. Forecasters I

The signs of forecast errors can be predicted using the difference between individuals' forecasts and the average of earlier forecasts of the same variable. It is possible to improve 115 forecasts without worsening any. It is difficult to reconcile this result with the rational expectations hypothesis, because the average of earlier forecasts is in the information set of the forecasters.

2. Forecasters 2
If forecasters compete in a tournament in which the forecaster with the smallest forecast error wins a prize, it is not an optimal strategy to minimize expected squared forecast errors. Instead, in Nash equilibrium, the distribution of forecasts corresponds to the distribution of outcomes. This prediction closely fits results of analysis of a very small data set.


4 Inequality and growth (won't go back there again so soon so don't ask)

The estimated partial correlation between income inequality and subsequent economic growth differs in an apparently significant way across studies. I think this is due to the huge debt crisis shock which hit many unequal countries at once in the 1980s. This is an interesting empirical question and a warning about the impluasible assumption that disturbances to growth are independent across countries.

5 Circular Spite

Evolutionary Economics Models are much simplified by the assumption of randome mixing and matching. It is considerably more difficult to handle models in which agents stay at a fixed position and compete with their neighbors. A very simple model of agents distributed on a circle gives a very different result for the possibility that suboptimal strategies are selected due to evolutionary spite.

6. The Breakdown Point of Instrumental Variables Estimators

I am aware of no research on the breakdown points of analogs of 2SLS, that is, of estimators which are consistent even when the explanatory variables are not weakly exogenous. The breakdown point is the fraction of arbitrarily chosen data points which can cause the estimator to take arbitrary value The challenge for a robust two stage estimator is to guarantee that, if the model estimated with the majority of the data is identified, so is the model estimated with all of the data. Maintaining the rank of the matrix of coefficients is clearly a much more difficult problem than preventing the estimates from taking any arbitrary value. Indeed it can be demonstrated that the breakdown point of any two stage estimator depends on the properties of the majority of the data and is typically bounded above by a value far below 50%. It is possible to describe an estimator which is consistent for jointly normal data in which the explanatory variables are endogenous and which has a breakdown point equal to this theoretical maximum.

7 An unbiased estimator of the variance covariance matrix of OLS estimates with Hetroskedastic disturbances

While standard Heteroskesastic consistent estimators of the variance covariance matrix are biased in small samples, an unbiased estimator is possible. This estimate is not useful for standard inference, but it is useful in testing the homogenteity hypotheisis in panel data.

8. A Diagnostic Statistic

OLS is highly sensitive to Outliers. If results are due to a few outliers, most residuals are reduced if the estimated coefficients are multiplied by a constant less than one. This count is a diagnostic statistic with little finite sample size distortion.

9 A modified information criterion for Stein class forecasting models.

Abstract if anyone is interested.

10. A new Pareto liberal paradox here


11. Who brings in the money and who spends time with the kids. Based on the first 3 wves of the ECHP, it appears that the time fathers spend with their children increases in the mother's income even after considering the amount of time the mother spends with the children. This supports bargaining models of the family as opposed to the standard Beckerian model.

1 comment:

  1. 1 and 2 look mighty interesting to me.

    -nikete

    ReplyDelete