0. Introduction

Models have been developed that reduce the risk of investing in the U.S. stock market, while increasing long-term returns. Algorithms that evaluate the market's price pattern over a given period were studied in relation to the market's subsequent performance. Various correlations were noted. The correlations were merged into a series of models that provide buy and sell signals.
 
 

The graph shows the annual risk-adjusted return (dividends excluded) of the NYSE and the three broad market models presented in this report over the 30 years from 1969 to 1998. Models #9 and #4 outperformed the NYSE Index and had a positive return in all 30 years. The aggressive Model #9A outperformed the NYSE Index and had a positive return in 29 of the 30 years, the exception being a 1.5% pre-dividend loss in 1985. This performance was accomplished without using short sales, in a market that declined in 9 of the 30 years.

This report:

1. outlines a philosophy for building effective predictive models;
2. documents the 30-year performance of three broad market models;
3. provides close-up views of model performance during bear markets;
4. presents the actual trading results for a market-neutral model; and
5. suggests potential applications for the developed techniques.

A central problem with developing predictive models based on past market performance is that the future market actions may not mimic the past. Financial situations never encountered by the market in the past, combined with new market-making technology, new modes of information communication and exchange, the interactions between derivatives trading and trading in other financial markets, and the impact of software-driven automatic trading can significantly alter market pricing mechanisms and market reaction characteristics, making mathematical methods that would have worked in the past fail miserably in the present.

Yet, in its essence the market always is driven in its long-term upward journey by human progress, and in its irrational spurts by the foibles of human nature--exuberance, anxiety, despair. This unchanging background leads one to posit that, despite continuous evolution of market mechanics, there remains an underlying fundamental nature that prevails and, we hope, can be detected, tracked, categorized, and used to aid us in understanding what the market is saying at a given moment, and most importantly, in predicting its most likely next move.

If there is a fundamental nature acting in the market, then it should be possible to invent models that work well under all conditions, in all markets. We would expect such models to return high profits in bullish market phases, and neatly sidestep the biggest losses in bearish phases. Importantly, we also expect these models to function well during choppy market consolidation periods, minimizing whipsaw losses when the market churns erratically.

The models presented here do precisely this over the 30 year data period (1969-1998) used to develop the models.
 
 

2. 30-Year Performance of Selected U.S. Stock Market Models

2.1 Investment Objective

The primary objective of the models is to minimize portfolio drawdowns. The secondary objective is to maximize profits.
 

2.2 Assumptions

Model Portfolio: In the plots and tables that follow, the "security" that is bought and sold is "the market", represented by the NYSE Index. In actual practice, an investor's diversified stock portfolio could be hedged using market index options or futures.

Dividends versus Cash: Because the market price data base does not include reinvested dividends, and the models switch between the stock market and "cash" (TBills, money market, etc.), an assumption was necessary on the rate of return of dividends versus short-term securities. The plots and tables presented here assume that an investment in cash yields a rate of return 3% higher than the market's dividend return throughout the 30 year test period.

Purchase/Sale Price: Purchases and sales are assumed to occur at the market's closing price.

Transaction Costs: Transaction costs are not included in the broad market models at present, though their effect has been evaluated.

Short Sales: Selling the market short is not a component of the models presented in this report.

Leverage: Market positions in excess of 100% are employed only in Model #9Aggr.

Derivatives Pricing: The purchase and sale prices for options or futures used to leverage market positions are assumed to be synchronized with the actual market index prices.
 

2.3 Model Performance Plots and Tables

The models are evaluated over the 30 year period starting in January 1969 in comparison with the market's overall performance (buy and hold strategy). Three models are highlighted in this report:

• Model #4: Very Conservative (~20% market exposure; no leverage; no short sales)
• Model #9: Moderately Conservative (~60% market exposure; no leverage; no short sales)
• Model #9Aggr: Aggressive (~110% market exposure; leveraged purchases; no short sales)

The parameters listed in the tables are calculated as follows:

• Total Gain = the percent change in the index price (NYSE or Model) over the entire time period.
• Annual Gain = the Total Gain annualized using the average number of trading days in a year.
• % Market Exposure = the average daily stock market investment percent over the time period [cash = 0%, fully invested = 100%, leveraged  (Model #9A only) = >100%].
• Volatility = average daily percent change in index price.
• Risk-Adjusted Return = annual return for NYSE; for Models: [annual return] * [NYSE volatility] / [Model volatility].
• Maximum Drawdown = the maximum percent difference between an index high and a subsequent index low during the time period.
• Maximum Days to Profit = the maximum number of trading days before the NYSE or Model reaches a new high during the time period (a + sign means the maximum was still counting as of 3/31/99).

Model #4 Performance vs. NYSE Index

Model #4 is a very conservative model, with an average market exposure of 20% over the past 30 years, scaling upward during more bullish periods like the past three years. The model has returned a profit every year in the test period (since 1969). The model can underperform the market during powerful bull runs, but usually catches up in the corrections. With a 30-year annual return of almost 15% on volatility less than 1/4 that of the market, the strategy provides an opportunity for risk-averse clients to experience stock-market returns.

Model #9 Performance vs. NYSE Index

Model #9 is moderately conservative, with an average market exposure of 60% over the past 30 years. The model has returned a profit every year in the test period (since 1969), outperforming the market in 24 of the 30 years. With a 30-year annual return of almost 23% on volatility just over half that of the market, the strategy provides an opportunity for substantially increased long-term portfolio growth at significantly reduced risk.

Model #9Aggr Performance vs. NYSE Index

Model #9Aggr is an aggressive strategy, but it maintains some conservative characteristics. The model employs leverage, but only when a large assortment of market conditions are in place. The model has an average market exposure of 94% over the past 30 years, a maximum drawdown less than half the market's '73-'74 drawdown, and volatility similar to the market. Yet, its 30 year return is almost 70% annually. Its performance is stellar in volatile markets (over 130% annual return in the past 3 years). The strategy is suitable for clients who seek very high returns for a portion of their portfolio.

3. Close-Up: Examples of Model Performance during Bear Markets

3.1 Performance of Model #4 in the 1973-74 Oil Crisis Bear Market

3.2 Performance of Model #9 in the 1987 Crash

3.3 Performance of Model #4 in the 1990-91 Gulf War Bear Market

3.4 Performance of Model #9Aggr in the 1998 Bear Market Correction

4. Application of Market-Neutral Trading Model (Actual Results)

The plot shows the actual performance (after paying broker commissions) of my trading versus an index constructed based on equal weightings of the stocks included in the model. The stock price index is adjusted to assume reinvested dividends. Applying a market-neutral method (half long, half short, using margin), the method earned 32% in one year trading low-volatility gas and electric utility stocks that returned 5% (including dividends).

The flattening of the profit curve at the right was due to two circumstances: 1) the market's volatility declined in late 1988, narrowing the spread between the stocks that were bought and sold short; 2) the size of my investment pool was reduced to the point where broker commissions were similar in size to my trading profits.

This strategy was to have been the initial product offering for my company, Innovative Strategies Investment Corporation. However, lacking a reputation, I was unable to attract clients. Since I didn't have the financial resources to continue on my own, I abandoned the strategy and dissolved the company.

The software and techniques used in the trading have been further developed and could readily be applied to today's market. Also, the cost of trading is substantially lower now than it was when I originally used the strategy, which would result in increased profits.
 
 
 
 
 

5. Potential Applications

5.1 Private Client Asset Preservation

A natural market for these techniques is high net worth individuals and families, for whom preservation of wealth is as important as its extension. The strategies would provide these clients with an effective means for hedging their portfolios against broad market downturns. Index options and/or futures could be used to fully hedge portfolios with minimal transaction costs, producing client returns similar to those presented in this report.
 

5.2 Corporate Mergers, Divestitures, etc.

The strategies could find application in mergers, new stock offerings, divestitures, and other situations where stability of corporate share value prior to the culmination of the deal is highly desirable. Here the strategies would provide a means for hedging against large drops in the share prices of the specific companies involved, to avoid a situation where altered market pricing makes a planned deal no longer financially viable.
 

5.3 Potential Application of Quantitative Methods to Global Economic and Business Decision-Making

The speed and data access capability of today's computers have altered the data analysis and modeling landscape utterly. This makes the potential for applying quantitative methods to the global marketplace unlimited. Economic and market patterns and anomalies can be detected, categorized, and modeled through thorough analysis of ever-growing data bases of economic and business information.

Historical economic and business records can be entered into data bases that are then examined and re-examined by a large variety of algorithms that search for patterns of change, cause and effect, correlations between indicators that occur simultaneously or in sequence. This type of analysis can be performed on virtually any data universe: the companies within a single industry; a country's stock market; its stock, bond, commodity, and currency markets; or, for the most complete picture, all available economic and market data for all countries can be integrated to form a complete global data base, to which quantitative analytic methods can be applied. This process could revolutionize economics, yielding new understanding of how the global economy works, how people and businesses, markets and governments, interact and influence one another. The effect on future global economic well-being due to improved policy-making would be immense.
 
 

6. Conclusion

Models have been developed that are believed to be of relevance to the investment banking community. The models provide a means for hedging investor portfolios to reduce risk while maintaining high return. It is believed that the models can be adapted, extended, and developed to yield models that would be an effective aid for economic and financial decision-making in the global marketplace.