A Brief History
Quantitative analysis of financial markets has a long and distinguished history that reaches back to the early 20th century with the publication of the groundbreaking paper "The Theory of Speculation" by Louis Bachelier in 1900. It was the first scientifically thorough study of statistical behaviour of stock prices. Unfortunately, his work did not receive the due attention it deserved. Bachelier was far ahead of his times. To a certain extent his misfortune was that his ideas lacked a practical catalyst: modern information technology. Ultimately the progress of modern finance and quantitative investing, in particular, has always been tied to the rapid advancement of computing power and development of comprehensive financial databases.
Finance took another leap forward in the early 1950s when Harry Markowitz introduced Modern Portfolio Theory (MPT), which auspiciously fell into the same timeframe as the invention of the first modern computer. Subsequently, William Sharpe extended MPT to the now famous Capital Asset Pricing Model (CAPM), upon which Fama & French later built their three-factor CAPM model, taking into account not only the overall market but also size (small versus large caps) and valuation (high P/B versus low P/B) as explanatory factors for stock returns.
To the early quant investors these market models provided the foundation for their search for alpha. But instead of looking at factors to explain historical returns they turned the models around using factors to forecast stock alpha. The classic multi-factor quant strategy was born.
Quant Investing-Where are we today?
Quant investing nowadays comes in many shapes and forms. Often it is still being misunderstood as the infamous "black box". The following provides a brief description of the most common quantitative strategies and talks about some of the important points to look out for when evaluating a quant investment process. Investors armed with the right knowledge will find that quant investing is anything but a "black box".
The classic multi-factor model is an example of a fundamentally driven quant strategy. The way it works is very similar to how a traditional stock analyst operates. In both approaches the basis of an investment decision is driven by a company's fundamental data (balance sheet, income statement, cash flow statement, financial ratios). The difference is that a stock analyst can deepen his analysis by interviewing management, talking to competitors or visiting production facilities. A quant model is limited to rating stocks based on the strength of numbers alone (EPS growth, P/E ratio, profit margin, etc). It can get some additional depth by tracking analysts' estimates, dispersion of estimates and rating changes. However, even though a quant model may not match an analyst in terms of depth of analysis, it excels in breadth. Thanks to today's abundance of computing power and extensive financial databases a quant model can monitor and analyse many more stocks in a systematic fashion than any human analyst ever could. Therefore a quant model's main strength lies in the disciplined process of applying a small edge repeatedly to find as many independent investment opportunities as possible. The resulting broad diversification increases the chances of achieving superior risk-adjusted returns.
The same principle applies to all quant strategies. Another important class are the purely statistically driven models which look at statistical properties of price action between two or more securities and derivatives. The standard example is pairs trading. Let's look at two companies that are related by being operational in the same business, say, two Japanese insurance companies. In the absence of major company-specific events it is quite likely that these two companies will trade in line and in the short term one company should not outperform the other significantly. Any such deviation should revert back to the mean. Statistics gives a quant manager the right tools to analyse whether two companies indeed trade in line and whether the odds for a profitable mean-reversion trade are favourable. This strategy is particularly computing power dependent. For example, let's look at all constituents of the Topix 500. We can form potentially 124,750 pairs. Even if only 1% of these pairs pass tests for mean reversion we still have to monitor more than 1,000 trades at any given time. Again, a small edge (reversion to the mean) is applied many times over (all combinations of suitable pairs) to yield superior risk-adjusted returns.
Lastly, let's examine another big branch of quant strategies: technically driven models. CTAs or systematic trading funds are the best examples in this category. The basic premise of these funds is that everything you need to know about a security (in this case mostly futures contracts) is in its price and volume traded. Usually several different models are being used, eg, market timing models for entry and exits, pattern recognition and momentum systems for trend analysis or mean reverting technicals for counter-trend trades. Any of these methods applied in isolation to only one security is not going to show impressive results. But just as before, applying all the various models on all the different futures contracts (typically we are talking about more than 50 futures markets) brings about the same diversification effect that allows a small edge to become a significant risk-adjusted return.
Strengths of quant investing
Widest possible investment universe. Computers do not mind crunching numbers, so anything that can be quantified and stored in a database can also be analysed as a potential investment input. This way quant investing can easily take advantage of opportunities in smaller under-researched companies and cover more companies in general.
Diversification. Every quant model in one way or another tries to exploit a small sustainable edge over many different (independent) trades. It can be shown mathematically that this will lead to a direct improvement in risk-adjusted returns (see Law of Active Management, Grinold & Kahn).
Disciplined process. Models are unemotional in their decision making process. Unlike human investors at times, models don't get "married" to an investment.
Weaknesses of quant investing
Depth of analysis is limited. Traditional analysts have an advantage in sourcing additional non-quantifiable information, eg, assessment of management quality, corporate actions and their impact, industry gossip, hands-on research as in factory visits, etc.
Strong dependence on data. Clearly a high quality financial database is necessary to run a credible quant research effort. Quant managers and investors alike should be aware of database limitations. For example, extreme data points: Quite often these points are entry errors of some sort. However, they could be valid data. Or backfilling of data: Toyota Motor Corp reported EPS for financial year 04/05 on 10 May 05. Many databases however will record the EPS number as of 31 March 05, the official financial year-end date.
Data mining and over-optimisation. If something looks too good to be true it often is. A model that may be fine tuned to deliver staggering results in a historical back-test is very likely to disappoint going forward. Look for a common sense approach in terms of portfolio composition.
Quant Investing - Performance Outlook
The sheer variety of quant strategies makes it impossible to give a one-fits-all answer but there are a few scenarios one can look out for that may threaten performance.
Fundamentally driven models work on the assumption that markets reward sensible investment decisions in the long run, eg, a cheap and profitable company should do better than an expensive and loss-making one given a reasonable time frame. The problem is that in a "junk rally" like the Internet bubble the exact opposite is the case. In this situation a fundamental model will go through a tough patch. Investors may want to check for components of the model that complement the pure fundamental approach, eg, market technicals like momentum.
Statistically driven models are sensitive event risks. Most stats models try to exploit small pricing inefficiencies. Hence, price gaps caused by corporate events or other unexpected news may cause big losses relative to the expected profit of a trade. Investors should look out for features like position limits, stop-loss limits and broad diversification.
CTAs / Systematic Trading Funds are broadly speaking trend-followers. Lack of clear macro themes will likely result in "whipsawing". Investors should find out to what extent the model can cope with non-trending markets.