Any active trader who has been trading any markets, be it Forex, Equities or Commodities, have come across the idea of system trading. Trading systems manifest a discipline where traders enter, exit and perhaps manage trades based on fixed and pre-determined rules without paying much attention to current market conditions. This approach of mechanical trading often relies on past or historical data to identify an “edge” that allows the system to yield supernormal profits.
How Mechanical Systems Work
Usually, statistical tools such as indicators and other technical analysis based tools such as trend lines, Fibonacci analysis and the concept of support & resistance levels are used to build mechanical trading systems. These systems are often coded into automated programs that can execute trades on their own without input from human traders. One of the biggest problems traders face when developing and trading automated trading systems is that the real-time performance of these systems often deviates from the demonstrated performance of the system in backtesting empirical or historical data sets.
The Underlying Structure of Free Market
In order to understand the root cause of this problem, much broader understanding regarding free market operations is required. Beside the exception of Initial Public Offering (IPO), all trades and transactions taken in the equity market, retail Forex exchanges and commodities exchanges represent a zero-sum market condition. Under the zero-sum game theory, these transactions don’t “create” any value, but the value just changes hands. Hence, one trader’s profit will constitute another trader’s loss. The significance of this theory is that market equilibrium is created by an agreement between two traders. The buyer values the instrument he is bidding for more than the price he is willing to pay and the seller values the offered price more than the asset or instrument he or she is willing to let go. Under this kind of circumstances, market always remains in a state of equilibrium. Therefore, mechanical trading systems that use oscillator type indicators such as stochastic and commodity channel index (CCI) to identify “overbought” or “oversold” levels don’t work. Simply, because market eliminates all overbought and oversold conditions as soon as each transaction is completed!
What Is the Meaning of Win Rate?
In reality what technical analysis based mechanical systems end up doing is, they compare the past market movements to the current situation and provide a probability measurement of the repetition of the past. This is called the win rate generated by the mechanical system. Because the win rate simply represents the probability of the price movement based on past data, there is no way to predict if a particular trade will be a winner or loser. It can only predict the winning percentage of a series of trades. Such as, by using this system, it expects to win 70 percent of the time over a 100 trades. However, the sequence of the winners vs. losers is not predictable. A system with a 70 percent win rate can end up losing 30 trades in a row then win 70 times or there could be a degree of randomness in winning or losing the trades.
Often we hear technical system traders say that because in past 5 years, when prices moved up this percentage, and indicator X’s values lined up with indicator Y’s values then price continued to go up to Z percentage more! In reality, they find that market condition is different in real-time compared to the past market conditions when the system was developed and backtested. As a result they end up losing money.
This happens because market is always evolving. The expectations, fears, greed and other mass psychologies are always different compared to past market conditions. There is no guarantee that the exact same market participants are trading now like they were trading for last few years when the system was built.
Past Data Can Not Predict Future Movement
Analyzing past data or backtesting only demonstrate a system’s viability during that time period. Let’s assume a trader is trying to develop a trading system by using data from 10 years period, 2001-2010. After spending sometime backtesting a simple moving average crossover, he found that on the daily time frame, 70 percent of the time, when moving average of 13 crossed moving averages of 50, prices continued to move up and reached a profit of 100 points. Now, when the trader will use the same moving average parameters to forward test his hypothesis during 2011-2012, he might end up only winning 40 percent of the time. The interesting part is that, he might end up winning only 20 percent of the time if he applied the same hypothesis on data from 1991-2000!
How to Effectively Develop a Trading System by Using Walk-Forward Analysis
When developers engage in developing trading strategies, they must understand the fact that no moment in the market is same, let alone a block of time consisting of a week or a year. However, there are practical remedies to safeguard trading strategies against this past bias that derive from backtesting. By using effective walk-forward analysis, often system developers get past the tendency to “curve fit” their strategies and get a real market edge that can generate profits in real-time.
Even with walk-forward analysis, system developers have to constantly take recent market conditions under consideration and put more weight towards the end of the data series compared to the distant past. In this way, last one month’s price data will carry more significant weight compared to data from five years ago.
System trading offers tremendous benefits to traders as they can just set and leave the system to generate profits without trading all day by getting stuck at the desk. It also allows traders to execute trades much accurately as automated systems are not prone to the “human error.” However, any trading system that relies only on backtesting results is destined to fail simply because the past does not represent the future. With walk-forward analysis enable traders to incorporate recent market behavior that offers better synchronization of the strategy to the market conditions. While backtesting is an essential tool to develop automated mechanical systems, it offers traders an incomplete approach to consistent profitability without walk-forward analysis.