Dec

10

There is something idempotent with the Knicks performance against Chicago and the 75 page paper by Hou, Xue, and Zhang.

They both start out so hopefully, and end up to me with a wimper. They suffer from look back effects, regression biases, part whole biases, multicomparison problems, and most of all basing a prediction on past results which contain many random factors.

The regression biases are overwhelming. How do the Knicks expect to win relying on a man like Smith whose shooting percentage is south of 30%? Why he did well the previous game, when the three percentage was almost 50%. Don't they realize that when they score that kind of % in a previous game, luck was involved to a large effect, and it is random, or negative serially correlated because the other team tries harder to defend against the threes and the Gallinari types like Smith are over confident.

Similarly in the Zhang studies, don't they realize that of course their results will appear significant if they base it upon already published results showing effects for the periods included in their study. Don't they realize that within a month, all the results of companies with different balance sheet characteristics are highly correlated and clustered, and that by the time they sort by dozens of variables with split after split they are left with few independent observations—certainly not enough to make meaningful significance. The companies in their various sorts don't change much from month to month, so they are measuring the performance of a small number of companies similar in style for say six months in the future…the tests, are certainly not enough to make any sort of meaningful predictions.

There is something to be said for their independent finding of change in assets divided by assets as a measure of past success and similarly for returns on equity. As far as I can see, however, they use a retrospective compustat file rather than the as is file and that makes all their results meaningless as companies with seemingly high returns on equity like Rimm often go from the black to the red and they appear to eliminate such companies from their comparison. Debt is not considered and with a retrospective file like Compustat, the value stocks will look great until they are delisted and not covered because of problems.

The study should have been performed with a given universe of large stocks with prospective data and data covering only the future years for their anomalies that were not already shown to have significant effects in past studies. Watching the Knicks hapless performance so typical of the Antoni led team of the past and reviewing this heroic but flawed study by Hou, Xue, and Zhang leaves relatively contemporaneousy leaves one with a certain sense of displeasure if not revulsion.

Alston Mabry writes:

Are papers like these read and digested and used by finance professionals? Who are these guys? This quote makes me think they are taking their own work quite seriously:

"Our work has important implications for academic research in finance and accounting. The qfactor model can be used as a new workhorse model of expected returns. Any new anomaly variable should be benchmarked against the q-factor model to see if the variable provides any incremental information above and beyond investment and ROE. More important, the vast anomalies literature in empirical finance and capital markets research in accounting should be reevaluated with the new expected-return benchmark provided by the q-factor model. Much work remains to be done."


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search