I'm looking into modeling the relationship between EPS announcement surprises with long-term returns (1 quarter to 3 years with intervals). I've based my current methodology off papers looking at the short term effect (example) but I think that the long time horizon will require a more comprehensive solution.
My ultimate goal is to be able to say with some degree of certainty whether or not beating or missing analysts' EPS estimates has a long term effect on the performance of a stock.
I've set up a regression with variables as follows:
I've defined EPS announcement surprises as
$$ \text{SUPRISE}_i=\dfrac{\text{EPS}_{actual,i}-\text{EPS}_{estimate,i}}{\text{EPS}_{actual,i}} $$
to create 2 variables for positive and negative surprises (POSSUPRISE and NEGSUPRISE)
Defined Returns as
$$ \text{RETURN}_t=\ln(\text{price}_{i+t})-\ln(\text{price}_i) $$
where $t$ is the final day of the time period I am analyzing
so my current regression looks like this
$$ \text{RETURN}_t = \beta_0 + \beta_p \text{POSSUPRISE}_{i}+\beta_n \text{NEGSUPRISE}_{i}+\epsilon_t $$
I've also done a regression with indicator variables for beating and missing estimates
I've run this over a sample set of 30 large cap stocks with EPS data from 1999-2009 and the appropriate pricing data and so far have had mixed results, I found some correlation between 2 year returns and large earnings surprises, but before I explore this question further, I want to make sure I'm going about it the right way
My questions are:
- Is a regression of individual instances the best way to analyze this problem? Should I use time series methods like VAR instead?
- What is the best way to incorporate broad market movement into the returns data? Should I just adjust the return variable to account for the return on an index over the time period as well or is there a better solution?
- Am I better off just considering the surprise variable or should I try to control for other variables in the model such as actual EPS, Market Cap, etc?