Of all investing disciplines, technical analysis should be the most amenable to quantitative research. I define quantitative research simply as the use of a computer and a database to identify investment factors that have historically been predictive of future excess returns (or returns above a market benchmark). Technical analysts rely on historical price and volume patterns to predict future price movement. Applying a computer to this task seems natural, and to the uninformed might seem easy.
In practice, however, much of technical analysis is an art, as is all else in the field of investment. A price pattern that works under one set of conditions may not work under another. What is easy for the eye and the brain to identify—a head and shoulders top, for example—can be very difficult to translate into computer code.
However, certain technical factors are relatively simple to model quantitatively. In this article, I’ll present a couple technical factors that performed well in the quantitative tests I conducted. The factors presented are simple, and should be nothing new to most technical analysts. What I hope to show, however, is that the excess returns provided by these technical factors can be made more consistent, and in many cases much higher (or lower, for short-sale strategies), by combining those technical factors with fundamental and valuation factors that also tested well quantitatively.
Sam Stovall, the chief investment strategist at Standard & Poor’s, introduced me to the phrase, “Fundamental analysis tells you what, but technical analysis tells you when.” With characteristic wit, Sam added “and quantitative analysis can tell you why.” I might put it slightly differently: quantitative analysis, done correctly, can provide investors with empirical evidence of the basic drivers underlying stock market returns. (But Sam’s quote is more memorable.)
John Murphy’s ground-breaking work on intermarket analysis enlarged and, perhaps, revitalized the field of technical analysis. Mr. Murphy’s work shows not only how the various financial markets are linked, but also how changes in one market (for example, a falling dollar) can foreshadow economic changes (e.g., rising commodity prices) that are significant in predicting changes in other markets (e.g., the bond and stock markets). I believe that these links between technical analysis and economic analysis warrant further exploration. My work certainly shows clearly that technical analysis and fundamental analysis (at the level of the microcosm—the individual company/stock), far from being mutually exclusive disciplines, work together in predicting future stock performance.
Nearly two years ago, I was asked to develop a series of quantitative stock selection models for the Equity Research department of Standard & Poor’s. In preparation for this project, we backtested over 1,200 different investment strategies, to determine which were predictive of future excess returns. (A backtest is simply a statistical look at historical data to determine if employing a given investment factor, such as selecting stocks with low P/E ratios, results in excess returns over time.)
My goal was to determine the basic factors that drive future stock market returns, from an empirical point of view, using only historical data as our raw material (balance sheet, income statement, cash fl ow statement, and pricing data). In short, I set out to create a quantitatively-drawn “road map” of the equity markets. To do our research, we used a sophisticated data-analysis program (Charter Oak Investment Systems’ Venues data engine) and Standard & Poor’s own Point in Time database, which contains over 20 years of “as originally reported” (unrestated) data for about 150 data items and 25,000 individual companies. This data-intensive approach to investment analysis yielded clear results. Certain strategies consistently outperformed the market over our two-decade test period, while others consistently underperformed. The results of this research will be published in an upcoming book, Quantitative Strategies for Achieving Alpha (McGraw-Hill, November 2008). In it, I present a wide variety of investment strategies that predict excess returns and show investors how to combine individual investment strategies into more complex screens and models that can be used to generate strong potential investment ideas, create quantitative portfolios, or simply better understand the market from a quantitative point of view.
In structuring our backtests, we kept in sight one basic principle: numbers can lie. If a backtest is not constructed carefully, or if too few years of data are used, backtest results will be unreliable. The researcher must consider different forms of statistical bias, such as look-ahead bias and survivorship bias (our database protected our tests from both). Returns must be calculated consistently—we used a stock’s annual price change plus dividends and cash-equivalent distributions of value (such as spinoffs). And a clear backtest universe must be defined: ours consists of the largest 2,200 stocks in our database selected by market capitalization with a minimum share price constraint ($2, to keep out volatile penny stocks).
Each test divides the companies in our backtest universe into quintiles (groups of five) based on their rank on one or more investment factors. For example, a P/E ratio test would put the 20% of companies with the lowest P/E ratios into the first (top) quintile, the next 20% into the second quintile, all the way down to the 20% of companies with the highest P/E ratios, which would be put into the bottom (fifth) quintile.
Portfolios representing each quintile are formed every quarter over our test period, and the holding period for each portfolio is 12-months. Returns for all portfolios in each quintile are then calculated, averaged over our 20-year test period, and compared to the average return over the same period for the overall universe. A strategy is said to have investment value if the top quintile significantly outperforms the Universe, the bottom quintile significantly underperforms, and outperformance/ underperformance is consistent over time. I like to use the idea of a mosaic to describe the results of our quantitative tests. A mosaic is a picture or pattern made by putting together many small colored tiles. In a real mosaic, each tile is meaningless when viewed alone, but when put together by an artist, a beautiful pattern emerges. In our investment mosaic, each “tile” is a strategy that has investment value (it consistently outperforms or underperforms the market) and is understood by the investor (we know why it works).
The second point is critical. Data mining—the search for correlations between items in a database—can uncover investment strategies that work fabulously during the test period, and fail to work thereafter. By basing the investment strategies we test on sound investment theory, we ensure that the results represent fundamental principles and tendencies in the investment markets and not statistical anomalies. When all the investment strategies presented in the book are put together, a mosaic emerges that shows quite clearly “what drives the market” from a quantitative point of view, and what characteristics to look for or to avoid in the companies and stocks in which we plan to invest.
So, what can investors learn from quantitative analysis? From a technical point of view, two strategies that worked well quantitatively were relative strength and 52-week price range. Our testing showed that using a seven-month period to calculate relative strength produced the best combination of excess returns and consistency over time, given our 12-month holding period. The 52-week price range strategy measures the proximity of a stock to its 52-week high or 52-week low. The formula is (current price – 52- week low)/(52-week high – 52-week low). The top quintile of the seven-month relative strength strategy generated average excess returns of 3.3% over our 20-year test period, and the bottom quintile generated negative excess returns of 3.4% (i.e., it underperformed our backtest universe by an average of 3.4%). However, the top quintile only outperformed the universe for 60% of the one-year periods tested, and showed the highest degree of outperformance from 1999 to 2000 (not an ideal characteristic for a quantitative test).
The bottom quintile was much more consistent, underperforming for 74% of the one-year periods tested.
The 52-week price range strategy had higher excess returns and was more consistent. The top quintile outperformed by an average of 4.3% over our 20-year test period, and did so for 75% of the 1-year periods tested. The bottom quintile underperformed by 3.9%, and did so for 79% of the 1-year periods tested. It was also less volatile than seven-month relative strength: the top quintile of 52-week price range had a Beta (versus our universe) of 1.0 and a standard deviation of returns of 0.18; the top quintile of seven-month relative strength had a Beta of 1.3 and a standard deviation of returns of 0.26.
However, these simple price momentum factors can be improved significantly by combining them with fundamental and valuation factors. Our testing employs a building block approach to quantitative analysis. The building block approach begins with seven major investment categories that our testing showed to be predictive quantitatively: profitability, valuation, cash fl ow, growth, capital allocation, price momentum, and red flags (risk). I call these categories the basics precisely because they are fundamental to achieving excess returns in the stock market.
Within these seven categories, we identified over 40 individual investment strategies that consistently produce excess returns. The seven-month relative strength and 52-week price range strategies, presented above, are two of these. I call these single-factor strategies building blocks. By combining a large number of building blocks, in simple two-factor tests, we learned which investment strategies work well together and which do not. These single-factor and two-factor tests form the heart of my book and help investors to form the mental mosaic of the stock market described above.
Combining two technical strategies, for example 52-week price range and seven-month relative strength, results in higher excess returns (about 9% for the top quintile) but in much higher volatility (a Beta of 1.50 for the top quintile). The strategy has extremely high excess returns from 1998 through 2000, but otherwise is not unusually impressive and has weak consistency.
However, combining a technical factor with a valuation factor results in high excess returns along with strong consistency and low volatility. For example, combining 52-week price range with free cash fl ow to price—a strong valuation factor—produced excess returns of 6.9% for the top quintile and 9.5% for the bottom quintile.
(Free cash fl ow equals cash from a company’s operating activities minus capital expenditures over the past 12 months.) The top quintile outperforms for 75% of 1-year periods and the bottom quintile underperforms for 82% of 1-year periods. Beta for the top quintile of 52-week price range and free cash fl ow to price drops signifi cantly, from 1.0 for 52-week price range alone, to 0.8 for the combined strategy.
Selecting stocks by valuation first further improves excess returns. (In our two factor tests,we select portfolios based only on the first factor first, and from the set that remains, we select the final portfolio based on the second factor. So, the first factor is always given more “weight” than the second in these tests.) For example, the free cash fl ow to price and seven-month relative strength strategy resulted in excess returns of 9.5% for the top quintile and 8.6% for the bottom; consistency is high and volatility (of the top quintile) is low. Valuation and price momentum make a natural pair, since valuation factors identify stocks that are “cheap,” while price momentum indicates that investors may see a catalyst (fundamental or otherwise) likely to turn the stock around. So-called “value” stocks often stay cheap for a long time. On the other hand, stocks with high price momentum sometimes sport very high valuations and are subject to significant declines on minor disappointments. The combination of valuation and price momentum ensures that a stock has both value and the potential for near-term price appreciation.
Another technical/fundamental combination that works well is price momentum and capital allocation. Capital allocation strategies look at what a company does with its excess cash. Positive capital allocation strategies, from a quantitative point of view, are share repurchases, debt reductions, dividend payments, and moderate levels of capital expenditures. Negative capital allocation strategies include large share and debt issuance, high levels of capital expenditures, and large business acquisitions.
Companies that significantly reduce their outstanding share count over a one-year period, for example, outperform. However, companies that significantly reduce shares and have strong relative strength have even higher excess returns. The top quintile of the one-year reduction in shares strategy generates average excess returns of 3.1%, and does so for 69% of one-year periods tested. However, the top quintile of the one-year reduction in shares and seven-month relative strength strategy generates excess returns of 5.8%, and does so for 70% of one-year periods.
Likewise, the bottom quintile of the one-year reduction in shares strategy, alone, generates average negative excess returns of 5.2%, and does so for 79% of the one-year periods tested. However, the bottom quintile of the one-year reduction in shares and seven-month relative strength strategy generates negative excess returns of 10.7%, and does so for 86% of one-year periods.
Although the valuation/price momentum combination presented above generates strong and consistent results, an even stronger combination can be formed by adding a profitability factor. The long screen I’ll present here (Table 1) takes advantage of three strong factors: operating profit to invested capital (profitability), free cash fl ow to price (valuation), and 52-week price range (price momentum). Operating profit is calculated as 12-month operating income minus depreciation expense. The complete screen is written as follows:
- operating profi t to invested capital > 20%
- free cash fl ow to price > 8%
- 52-week price range > 60%
Table 1: Quantitative Screen
The values for this screen were taken from the appendix of my book, which provides the average portfolio values over time for each of the building blocks presented in the book. Screen values will vary depending on market and economic conditions. For example, average values for the top quintile of the 52-week price range strategy have varied from a low of 46% in 1987 to a high of 96% in 2003. I chose 60%, simply because using a higher value in the midst of a severe primary downtrend (as of October 2008) meant that very few stocks would pass the screen. Note that data for this screen is as of September 30, 2008. An 18-year backtest of this screen yields the following results: Compound annual growth for the strategy of 18.9% versus a CAGR of 10.3% for the backtest universe, outperformance versus the universe for 94% of 1-year periods tested, a maximum loss over any 1-year period of 10%, and a Beta versus the universe of 0.90. The accompanying table shows select stocks chosen from this screen, which generated 25 companies using data as of September 2008. I narrowed the screen primarily by choosing stocks that looked best from a technical point of view.
This screen is an example of one major conclusion of our work: fundamentals matter, valuations matter, and technicals matter. The investor looking to achieve strong stock market returns over a six-month to one and one-half year investment horizon would do well to consider all three of these factors.
Another important conclusion is that quantitative analysis, qualitative analysis, and technical analysis form mutually complementary disciplines—investors who learn the lessons taught by each are apt to increase their ability to make money consistently in stocks.