|
|
NUMERIC INVESTORS
CASE STUDY
by
Bruce Egsieker
Numeric Investors is a Portfolio Fund Management Company, located in Cambridge Mass.
Lang Wheeler is Numeric's founder and C.E.O. His partners are John Bogle and Mark Engerman.
Numeric Investors was founded in 1989 by Langdon Wheeler, who had developed the original version of the firm's earnings estimate revision model. This case study is about computer modeling of companies financial information and stock market pricing. Lang had studied the effect of analyst's earnings estimates on the stock prices of 300 companies. The analysis and ideas he gained from this exercise would eventually become Numeric's Momentum Model.
By 1985 Lang had built a successful model. Numeric manages tax-exempt accounts for pension funds, for endowments, offshore funds and for three mutual funds which it started in June of 1996. The mutuals are named: Numeric N/I Micro Cap, Numeric Growth, and Numeric Growth and Value Funds. In the spring of 1997 the company had $3.0 billion dollars in assets across seven different product categories and had 21 employees. Numeric, for management of the money, operated under a variety of fee arrangements. They preferred to be compensated through performance fees, rather than flat fees.
Numeric has two strategies as to investment that it follows. These are:
Strategy One...For the long positions, Numeric would buy "good" stocks and for the short positions, it would sell short "bad" stocks. In this way the firm could exploit its ability to predict losers as well as winners, a so called "double alpha" strategy. For Example: Buying "good" energy stocks and selling short an equal dollar value of "bad" energy stocks would eliminate most of the portfolio's energy sector exposure.
Strategy Two...Equitization consisted simply of the long/short market-neutral index portfolio overlaid with stock-index futures equals to 100% of the portfolio's net value. Most of Numeric's long/short accounts were equitized.
They saw several advantages in long/short investing. This allowed them to exploit negative as well as positive information and these long/short portfolios could have any sectors weights. So, long-only strategies tended to bundle diversification along with stock selection whereas long/short strategies could be managed more easily for pure stock selection. Long/short strategies also could save costs.
Numeric's stock selection Models were of two dominate kinds. The models evaluated 2,400 stocks and gives them a "momentum" score and a "value" score. Each score range is from +3 to -3 best to worse.
I. Momentum Model...The Momentum Model is based on momentum investing which is based on a belief that security analysts tended to underreact to expected changes in corporate earnings and investors tended to underreact too. The strategy was to get in early and ride the stock, either up or down, until its earnings revisions ran out of momentum. Within the Momentum Model they used two approaches to forecasting future earnings estimate revisions. 1) an approach based on past changes in analyst's estimates. These small revisions would then encourage other analysts to update their estimates. This would lead to earnings estimate leapfrogging and a herd effect. They captured this effect in its "Estrend" or earnings estimate "trend" model. Candidates with large increases in analysts' estimates were candidates for buying and companies with large decreases in analysts' estimates were candidates for selling short. 2) the second approach to forecasting future changes in analysts' earnings estimates was based on earnings surprises...company announcements of quarterly earnings that were significantly different from the consensus of analysts' expectations. A stock price increase immediately following the announcement of a positive earnings surprise usually would signal that the earnings were indicative of good future outcomes for the company. However, a stock decline might signal perhaps that, while the last quarter may have been better than expected, other information released in the announcement was indicating that the improved earnings were unlikely to be sustained. So for a given stock, Numeric's models determined both an Estrend score and an earnings surprise score. These scores were then combined in a weighted average to form a stock's momentum score. Earnings surprise tended to be the most effective during the days immediately following an announcement and Estrend tended to be more effective after some time had passed.
II. Value Model...The basic premise of the fair value model was that stock prices were more volatile than the underlying company fundamentals. This meant that prices would exhibit mean-reverting behavior, which could be forecast based on company fundamentals. This model used a cross-sectional regression analysis to determine the statistical relationship between certain characteristics of a company and its stock price. Variables like: earnings, book value, earnings growth, earnings quality, economic sector membership, and analyst coverage. A regression equation is used here to calculate a "fitted price" or "fair price" for each company based on its characteristics. So a stock could be cheap or expensive if it was above or below the "fitted price."
Numeric found that the Momentum and Value Models complemented one another. They did eliminate such stocks such as biotechnology and gold mining stocks as they were not sensitive to either model. Momentum had the most skill and value but the least skill among technology stocks. Momentum had the least skill among utility and basic industry stocks. Value had the most skill among financial and basic industry stocks.
Numeric's strategies involve considerable portfolio turnover. The range is from 160-200% per year. During the the year 1997 Numeric traded 640 million shares worth $22 billion dollars. During 1998 and 1999 they traded over 700 million shares each year. Numeric made the decision for now to close many of its products to new investors. Therefore, they will only manage for their current customers. They are intentionally staying small. It's mission is: 1)be good, not big 2) focus on returns, not marketing, and 3) rely on research and improve model continuously.
Case study question to be answered and forwarded to Mr. E. is:
1) What methods is Numeric using to make money for its customers. What are they doing to obtain asset growth for their clients and maintaining that growth? (think carefully here and smile)