This graph shows the performance of a group of stock portfolios, each selected based on a set of criteria intended to identify undervalued companies. Each point on the graph represents the average performance of a group of portfolios. Each portfolio consists of stocks selected automatically based on a set of parameters dependent on the portfolio's position in the graph. The height of the surface at each point indicates the percentage by which the portfolios at that point outperformed the S&P 500 index. This particular graph represents the performance of over 26,000 portfolios in total.

In the case of the above graph, all portfolios identified by the trading strategy being evaluated outperformed the S&P 500, by factors ranging from 5% to 40%. (Note that the higher returns are predicated on short holding periods, which for various reasons including volatility and tax-efficiency may not be achievable in practice.)

Graphs such as these can be used to optimize a trading strategy on an ongoing basis, reacting to changes in market conditions. Research and experience have shown that the overall shapes of these surfaces remain consistent over long periods of time, so while it is never possible to ensure that future performance will be as expected, it is possible to continually adjust portfolios so as to be best positioned to outperform the market.

This technology is currently being successfully used by an offshore hedge fund with excellent results, currently outperforming relevant market indices by 30%, with an absolute annualized return of 83%. These results are significant, because the average fund does not consistently outperform the market. In fact, according to Standard & Poors, "only 14 diversified equity funds beat the [S&P 500] index's annual return of 22.98% for the five years ending with November, 1998."

Why do such techniques work? A variety of academic and industry research, combined with actual experience, has shown that on average, investors tend to "overreact" in the short term to good news or bad news, without taking into account other contextual factors. This leads to price distortions which, analyzed correctly, can be detected and exploited. One successful approach which arises from this is "value" investing, which people like Warren Buffett have used to impressive effect. By buying stocks which appear to be undervalued, you have a greater chance of performing better than the market. These results have been confirmed repeatedly in theoretical and real world studies.

Most value investing approaches either rely on examination of a company's actual operations, which involves a great deal of subjectivity and company-specific expertise (Buffett's approach), or are based on fairly simplistic measures like p/e ratios, which are of questionable relevance under current market conditions. An alternative approach involves analyzing past share performance and other statistics, and using statistical techniques to identify parameters which can be used to select portfolios of stocks which, as a group, are highly likely to outperform the market. Care must be taken to avoid falling into the classic data mining trap, in which patterns are discovered in the data which have no predictive value. However, with careful analysis, such traps can be avoided. The result can be a trading strategy which is capable of consistently outperforming market benchmarks. Such strategies are now being used by some of the largest fund management companies.

For the application described here, high-performance software in Scheme and C++ was developed to run the portfolio simulations. For each set of parameters tested, hundreds of thousands of portfolio simulations would be run over 12 years of historical data, to obtain a set of optimal portfolio selection parameters. These parameters can then be used to automatically select stock portfolios. The software is packaged in the form of ActiveX objects, which makes it possible for a user to experiment with different combinations of parameters from within a familiar environment, such as an Excel spreadsheet. The final graph displayed above was produced by the Open Visualization Data Explorer, an advanced scientific visualization product released as open source software by IBM Research.

For further information on:
- Financial analysis and investment methodology: contact Chris J. Muller at chrism@appsolutions.com
- Software development: contact Anton van Straaten at anton@appsolutions.com