Encyclopedia of Financial Models, Volume II by Fabozzi Frank J.;

Author:Fabozzi, Frank J.;
Language: eng
Format: epub
ISBN: 4789969
Publisher: John Wiley & Sons, Incorporated
Published: 2012-09-28T00:00:00+00:00

APPLICATION OF CART IN STOCK SELECTION

In this section, we provide a detailed example of the CART algorithm as applied to the problem of identifying profitable stocks. This example was specifically chosen so as to provide a contrast with the vast majority of the linear modeling techniques used by financial practitioners. The model was built with monthly stock data from December 1986 to August 2010 covering all liquid stocks listed on the North American equity markets but excluding financial stocks because they would require their own specific model.1 The number of total observations is 279,188 (or 980 stocks per month on average).

At the end of each month, forward total stock returns (price return plus dividends) were calculated. Using the median return of all sample companies in the same period as a proxy of the market return, the excess returns were then computed as the total returns minus the market returns.

A broad spectrum of company valuation and quality-based characteristics, as well as measures of investor sentiment such as price momentum and earnings revisions were selected as reported in Table 1. Instead of using raw values, we use rank orders in order to improve the robustness of the analyses. At each month, the rank order for each variable was computed by first ranking n stocks according to the corresponding variable value, and then dividing the rank by n to scale it between 0 and 1. Furthermore, in order to overcome the high correlation among some of the explanatory variables, nine composite factors were promoted as potential explanatory variables, which were constructed as an equally weighted average of multiple variables as described in Table 1.

Table 1 Input Variables

Composite factor Description

Value

(VAL) An equally weighted average of value metrics including dividends to price, cash flow to price, sales to price, and book to price.

Profitability

(PROF) An equally weighted average of profitability terms including return on equity, cash return on equity, pretax margins, and asset turnover.

Leverage

(LEVERAGE) An equally weighted average of financial strength terms including debt to equity and debt to market cap.

Debt Service

(DEBT.SERVICE) An equally weighted average of debt sustainability measures including interest cover and free cash flow to debt.

Momentum

(MOM) An equally weighted average of momentum terms measured over various time horizons including 6 months and 12 months.

Stability

(STAB) A composite term that captures the volatility in earnings, sales, and cash flows over the previous 5 years.

Historic Growth

(HIST.GROWTH) An equally weighted average of 3-year historic growth in earnings, sales, and cash flow.

Forward Growth

(FWD.GROWTH) An equally weighted average of I/B/E/S forecasted earnings growth expectation for FY1 and FY2.

Earnings Revisions

(EREV) An equally weighted average of the 3-month change in I/B/E/S forecasted earnings expectations for FY1 and FY2.

We built a classification tree with the purpose of predicting subsequent stock performance. Stocks were sorted into two groups, “outperformers” for those with positive excess returns and “underperformers” for the remainder. The induced categorical variable was then used as the dependent variable in the subsequent modeling process. One of the benefits of working with categorical responses instead of raw returns lies in the fact that it alleviates the impact of extreme returns, which may have multiple causes.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.