Artificial Intelligence May Revamp Future Trading System Development
The myriad ways that Wall Street has devised to make enormous sums from trading activities can at times astound the mind. The past few decades have witnessed the unbridled advance of quantitative trading, producing millions in revenue from the efforts of geniuses in mathematics and physics with the aid of complex financial trading algorithms. It has not been a smooth road along the way. There have been colossal failures and near meltdowns on a global scale, especially when market conditions stray beyond two standard deviations from the mean.
The successes, however, have far outweighed the failures such that the growth in hedge fund activity has grown in lockstep with double-digit returns posted by the leaders in this field. Success does breed copycats, and with more fund managers focused on similar "stat arb" trading strategies, return spreads have tightened, and lucrative arbitrage opportunities have become harder to find.
At present, most firms are conducting post-mortems on their proprietary "black boxes", trying to find the right mix of adjustments to stay ahead of aberrant market conditions. The general problem is that when every piece of software is arriving at the same conclusion, the market becomes flooded with sellers with no buyers in sight. The precipitous falls that follow require a new "safety net" to prevent their reoccurrence in the future.
One subset of the programming universe has been focusing on an area of artificial intelligence for potential solutions. Genetic Algorithms ("GA") and Genetic Programming ("GP") are new areas of endeavor that have evolved over the past twenty years and may offer a new and improved "self-adapting" approach to programmed trading systems. Although most of the work on the topic has been a series of academic papers and books, one recent popular guide on the topic cites a variety of real-world applications that include: "curve fitting, data modeling, symbolic regression, image analysis, signal processing, financial trading, time series prediction, economic modeling and more."
The world of academia is firmly attached to the Efficient Market Hypothesis, which states that current prices reflect all available public information and investors and markets act rationally. Technical Analysis is roundly criticized as a set of rules that are randomly lucky for various reasons. If markets were truly this "perfect", then prices would not fluctuate. Volatility, the reason for active trading in the first place, creates the opportunities that technical indicators and pattern recognition attempt to leverage.
"GP" offers a pathway for automated trading plan development, and there have been some successes documented from the stock market to the forex market. If a basic decision-tree can be designed, GP program sets can actually find solutions in an iterative manner by effectively "pruning" the tree and trying out new directions in a similar fashion as biologic evolution and the survival of the fittest. There must be a "fitness function" that evaluates incremental modifications, and the "art" involves creating a robust set of conditions that the programs can illicit to mimic market situations.
Tests on currency pair data from forex brokers have attempted to optimize a decision-making process where a large amount of variables present an even larger set of variable combinations to ponder as the search for a fitter "generation" of code proceeds. As various "mutations" are evaluated, the result is a trading system for a specific currency pair over a specific timeframe. One experimenter arrived at a GP solution that achieved profitability by alternately using Stochastics and RSI indicators to detect trends and reversals.
Genetic Programming may be the wave of the future. The results may seem counter-intuitive, but "evolution" does work in mysterious ways.