Handling Unprofitable Companies in P/E Regression Models

Summary

The document considers how to include valuation data in a regression when some companies have no meaningful positive price-to-earnings ratio because they are unprofitable. It raises two possible treatments: assigning those firms an artificially high P/E, or adding a profitability indicator and allowing P/E to contribute only for profitable firms.

The accepted response proposes using earnings-to-price, the reciprocal of P/E, to keep the ordering consistent across positive and negative earnings. This offers a compact alternative to imposing an arbitrary large value or separating the cases with a dummy variable. The note does not compare these choices empirically or discuss how negative earnings, zero earnings, outliers, or model interpretation affect estimation. Its recommendation is therefore a brief modeling suggestion rather than a complete treatment of valuation variables.

Key ideas

• Unprofitable firms do not have an ordinary positive P/E ratio, which complicates regression inputs.
• Assigning an arbitrary high P/E to loss-making firms introduces a chosen threshold into the data.
• A profitability indicator can distinguish profitable firms while restricting the P/E effect to them.
• The response suggests earnings-to-price as a reciprocal measure with consistent ordering.

Tags

# PE ratios in regression models - How to deal with unprofitable companies?


# PE ratios in regression models - How to deal with unprofitable companies?












First of all the following is for personal project, not to actually trade, so I'm under no illusion that a simple regression is going to make me money.

Pretty simple question: Suppose you're fitting a simple regression model with the most recent PE ratio being one factor.

The problem obviously arises that if a company is unprofitable its PE ratio is undefined (or, if calculated using the normal definition, leads to inconsistent ordering due to negative PE).

I'm curious how one normally deals with this?

My two ideas would be to either:

- Set unprofitable companies to have a large PE ratio $M>0$ such that $M$ is larger than the PE of any profitable company in the dataset

- Introduce a dummy variable $Profit$ that is $1$ if the company is profitable, $0$ otherwise. And replace the $PE$ variable in the regression with

$$PE:Profit + Profit$$

Since this includes an intercept to distinguish profitable and unprofitable companies, and also includes a PE slope for companies where PE is defined.

General thoughts on this issue are appreciated.

## Answer by Freelunch (score 2, accepted)

https://quant.stackexchange.com/a/42650

Use the $EP$ ratio instead, with $EP = \frac{1}{PE}$. Now your ordering will be consistent.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.