A “Quant” Approach to Predicting Revenue Growth

Lecture Center 13

The primary task for an active equity investor is to identify stocks with positive expected future returns. In general, this is a very difficult task, as stock returns are a function of many inputs, including size, value, and profitability. To predict company profitability, it is necessary to develop models for company revenue and company costs - as they are independent functions. This study attempts to model the former, by developing a two-stage algorithm to estimate the probability that a company will report quarterly revenue growth. This is done using a combination of Time Series and Bayesian statistical techniques. In the first stage, a Logistic Auto-Regressive Moving Average (LARMA) model is used to estimate a prior probability that a company will report revenue growth, based on historical quarterly data from the Compustat database. A limitation of this prior estimate is that it cannot incorporate data that is more recent than the last quarterly report (three months ago), during which the probability of revenue growth may have changed. There is an attempt to remedy this problem in the second stage of the algorithm, which utilizes Bayesian conditioning to update our prior estimate based on the proportion of similar companies which reported revenue growth in the past month. This study aims to identify whether this Bayesian conditioning significantly improves the accuracy of our prediction. This project is currently a work-in-progress, though sufficient results are expected by the time of the conference.