Dear Stata list members
Advance apologies for asking a naive question. I conducted an analysis to check whether firm-level working capital prior to the COVID-19 crisis(i.e. working capital level ex-ante crisis) is associated with the stock performance during the crisis period(crisis period, for instance, 1st quarter of the year 2020). My idea was motivated by recent media reports and anecdotes citing that firms will be facing a low demand hence those firms with excess inventory, more receivables (inventory, receivables and payables etc are the constituents of working capital) will be hard hit by the outbreak of COVID-19 (alternative logic says excess inventories will be good for during crisis times). Of course, the working capital level will be changing but still, there will be firms which may have stocked more inventories vis-à-vis with their peers. How do market value such firms during the crisis time based on their working capital measured before the crisis. To operationalize this, I ran an OLS with the dependent variable, “return” measured as Log of (Price at 31/03/2020/ Price at 01/01/2020), and following independent variables,- inventory scaled by total assets accounts receivables scaled by total assets, accounts payables scaled by total assets, cash holdings scaled by total assets, Ln(total assets) and industry dummies. All the independent variables are measured as of 2019 March.
Thus, I have a cross-sectional data where each firm will have one observation as dependent variable(returns) with independent variables
Results: I ran the above regressions and I got a poor model (by F value and adjusted R^2) in which none of the coefficients was significant.
I repeatedly did this with different returns considering the different period and none worked.
I have read that if none of the independent variables are statistically significant, the overall F-test is also not statistically significant. Also, F stat is a joint test which tests whether the linear regression model provides a better fit to the data than a model that contains no independent variables. This was true since my intercept was having a higher magnitude and was statistically significant
Before giving up this idea, I have a few doubts
Advance apologies for asking a naive question. I conducted an analysis to check whether firm-level working capital prior to the COVID-19 crisis(i.e. working capital level ex-ante crisis) is associated with the stock performance during the crisis period(crisis period, for instance, 1st quarter of the year 2020). My idea was motivated by recent media reports and anecdotes citing that firms will be facing a low demand hence those firms with excess inventory, more receivables (inventory, receivables and payables etc are the constituents of working capital) will be hard hit by the outbreak of COVID-19 (alternative logic says excess inventories will be good for during crisis times). Of course, the working capital level will be changing but still, there will be firms which may have stocked more inventories vis-à-vis with their peers. How do market value such firms during the crisis time based on their working capital measured before the crisis. To operationalize this, I ran an OLS with the dependent variable, “return” measured as Log of (Price at 31/03/2020/ Price at 01/01/2020), and following independent variables,- inventory scaled by total assets accounts receivables scaled by total assets, accounts payables scaled by total assets, cash holdings scaled by total assets, Ln(total assets) and industry dummies. All the independent variables are measured as of 2019 March.
Thus, I have a cross-sectional data where each firm will have one observation as dependent variable(returns) with independent variables
Code:
Log returns for each firm=Constant+ Beta*receivables of each firm + Beta*payables of each firm……+Beta*ln(total asssets) + Industry dummies
I repeatedly did this with different returns considering the different period and none worked.
I have read that if none of the independent variables are statistically significant, the overall F-test is also not statistically significant. Also, F stat is a joint test which tests whether the linear regression model provides a better fit to the data than a model that contains no independent variables. This was true since my intercept was having a higher magnitude and was statistically significant
Before giving up this idea, I have a few doubts
- Why even known predictors of stock returns like Log(total assets) didn’t show significance?
- Was my model a wrong one (using OLS, wrong measure of returns etc) even though same models got results in many working papers
- Market reaction has nothing to do with my independent variables (as my F statistics indicated)
Comment