wilmott.com

Hello all.I am in a real tight spot at the moment since i get information which at the moment i dont trust from my supervisor concerning the suitability of a model on profit warnings .I regress Abnormal volume, on returns plus some dummy variables that contain information about the nature of the profit warning.Vi,t = aRi,t+ DummiesI was told that this regression doesnt hold econometrically, however i am puzzled since i find similar work on scientific papers (Bamber et al.). Reasoning behing this claim that my regression is "wrong", is that Abnormal volume and Returns are endogenous variables and i cant use them in the same regression (!!!???). As far as i am concerned A.Schleifer used them.Anyhow all i need to know if there are any associated problems in using them as in the above equation. I was proposed to use them but not to run a simple cross sectional regression. Instead i should use simultaneous equations of the following :Vi,t= aRi,t + DummiesRi,t= bVi,t + DummiesExcuse any ignorance that might be evident in my questions, but i am a mathematician by trade doing a phd in behavioral finance.YoursIoannis

Intuitively you would have endogeneity problems between those two variables. So use Durbin-Wu-Hausman to test for this, and then run 2SLS/IV.

Thank you very much,is that the Durbin Watson test you are referring to?What is the 2SLS/IV?Again excuse my ignorance

2SLS : two-stage least squaresIV: instrumental variablesI think the Hamilton's book on time series has implementations of them.Kyriakos

you are right Rez, Hamilton explain the problem in details and gives examples.

Ok,So far i was running a simple cross sectional regression for this equation,I have read in some papers that Returns and Ab.Volume are positively correlated. But what i am concerned with not if they are correlated or not in general, but if one affects the other based on arrival of new information (announcements).You believe that i will have problems using a simple OLS crossectional regression on this one? I am doing a 3000observations event study (3000 companies in 3000 different dates respectively)I will check on this Hamilton book (time series analysis).I am looking forward to something directly applicable for computing this instead of an analytical solution (ie to directly solve it with 2SLS on my computer.)Any further advice would be more than wellcome, and many thanks for the already given advices.BTW. I was proposed to alter my regression by REMOVING the Returns and stubstituting them with LAGGED Volume (-1day), any comments on this?I thought that a regression like R= aV+b, or V=cR+d was essential for finance... can we just remove R like this?????Yannis

Ioannis :One thing is not clear to me: What is the object of your regression. Are you finding how volumes move returns or how returns move the volume. Well, one thing with just using volume [instead of change from previous day] is that volumes are not covariance stationary. Hence, you got to use change in volume from the previous day or time step. Its the similar reason as to why you would use returns instead of prices.If you use a change in volume ( as dependent variable), then ,yes the equation is econometrically correct. Otherise, as someone said you have to use the lagged variable as independent variable (on the right side) and the returns on the left side as dependent variable. I would actually use log of the volume instead of volume because the variation in the log wont be so much as in the actual volume. Another thing you may have to worry about is , in case you use the data for many time periods, about the cross sectional variation of the volume and the returns. That is change from one month to another month of the stocks. In which case, you have to normalize the data for each month or use a Fama-Macbeth kind of regresion. That is regress for each month. And then average the loadings of all the months to get a final estimate. Hope that helps.

Ok here is the dealI am regressing abnormal volume (left handside) on Returns and some dummie variables (which interest me) like in the following equationABVi,t = aRi,t + bDummiesi= profit warning announcement for i company , t= time respective to the announcement.So i have a list of 3000 companies/announcements like the followingIMB , Bad news on 10-October 2000 Vibm, RibmDell, Bad news on 12-October 2000 Vdell, RdellAMD, Bad news on 15-October 2000 Vamd, Ramd...etc.The object of my regression is how those Dummies affect Abnormal Volume, and not how returns affect volume or vice versa. Those dummies basically describe the nature of each announcementSo, so far, i used a simple OLS regression on the above regression model, and got some nice results, concerning the effect of the Dummies on Volume. However i was (wrongly?) told that Volume and Returns are endogenous variables and that this might cause a problem in my regression. However i am not concerned on how Volume is affected by returns, but basically only how volume is affected by the Dummies which represent the type of news on each announcement.Thanks again for all the helpIoannis