## Price Action Lab Blog

#### Premium Market Analysis

Regardless of your market and timeframe, profitable trading requires that the profit expectation is greater than zero. This is a requirement of profitable trading and not the secret as some book authors argue. The secret is finding a sequence of market entry and exit points that will achieve a sufficiently large expectation throughout the trading period at an acceptable maximum drawdown level.

Trade profit expectation, a.k.a expectancy, is a very illusive concept. The profit expectation E[T] is calculated in the trading literature using the averages of winning and losing trades as follows:

E[T] = w × avgW – (1-w) × avgL                   (1)

where w is the win fraction, avgW is the average winning trade and avgL is the average losing trade (in absolute value).

However, the above definition is problematic in many respects. As I have shown in this post, the right hand of equation (1) equals the average trade:

AvgT = (∑W – ∑L)/ N                                        (2)

Where ∑W is the sum of winning trades, ∑L is the sum of losing trades and N is the number of trades.

The first problem is that the expectation E[T] is approximately equal to the average trade AvgT only for sufficiently large samples (Papoulis, p. 138). But what is a sufficiently large sample? That depends on many factors and varies.

The second even more important problem is that E[T] is not stationary in general but can vary as a function of trade count N. If E[T] varies, we want the value to stay positive, of course, throughout the course of trading activity but also remain large enough so that it results in a satisfactory return above what we get from “risk-free” interest rate. We also want the variance to be small so that the resulting equity curve is as smooth as possible and without large swings.

Why most traders fail to maintain a sufficiently large trading expectation?

(1) It is quite unlikely that discretionary trading can maintain sufficiently high expectation for extended periods of time. This is the second most important reason that most trades fail. (The first and most important is account undercapitalization.)  The outcome of decisions based on discretion will approach a win rate of 50% over time due to the law of large numbers. Traders will be able to stay profitable only if they can make more than they lose on the average. This is not easy to accomplish. A careful analysis of the markets shows that low risk/reward ratios are hard to realize and maintain. But some will succeed due to either luck or skill but remain a small percentage of the total.

(2) Traders using technical analysis have no chance of staying profitable over the longer-term because this type of analysis easily leads to compulsive actions triggered by subjectivity and wishful thinking. There are some that have developed an edge using this type of analysis but they have invested the amount of time and effort that is beyond the norm. In my opinion, 95% of technical analysis users cannot maintain a positive expectation.

(3) System traders have a chance because this approach of generating expected values minimizes emotions and/or illusions. However there are some problems with some approaches followed that can lead to performance similar to that of discretionary or technical analysis trading. Specifically, the following approaches to developing a trading system can lead to negative expectations:

(a) Systems based on technical analysis indicators, chart patterns or naive candle stick analysis (out of context). There is basically no difference between a system trader using an automated technical analysis program to find double bottoms, for example, and someone who does it manually. They are both destined to fail unless they have managed to extract some kind of an edge from this type of analysis

(b) Systems developed automatically using genetic optimization  or programming, neural networks, walk-forward optimizers that incorporate indicators or algorithms with adjustable parameters. These type of systems involve Type-I curve fitting. They looks good on paper but the expectation turns negative as soon as the underline market conditions change. Some of these programs do not worth the amount of memory they reserve on a computer, yet many traders use them because of hype.

Methods that seem to work under certain conditions

(A) Very high frequency trading. This style of trading is beyond the reach of retail traders due to its complexity and high upfront cost. It is a proven high expectation generating process and this can be justified mathematically. It has nothing to do with front-running or other misconceptions of the public. The proof is beyond the scope of the post but intuitively, these traders profit from providing liquidity to the market.

(B) Other proprietary methods that are fairly unconventional but constitute an edge and are often based on alternative data sets.

(C) Parameter-less systems of Type-III curve fit. An example of a system that generates parameter-less systems based on price action only is  DLPAL, the software I have developed. This program allows finding strategies with a large enough sample of trades and high reward/risk. Some people are confused and think that each and every price action signal must have a large sample. This is not true in the case of such systems because price action is similar action and only differs in details. As a result, one looks at the sample size of a system that consists of many price action patterns/signals, for example. There are of course issues here, for example determining the significance of out of sample tests. The method is not trivial.

(D) Trading price action based on support & resistance, swing high/low, etc. This is also a parameter-less method but unless some strict guidelines are used, it can result to compulsive actions and negative expectation. I am not aware of any concrete set of rules for this type of trading but I am sure some have developed then and use them.

More information and detailed examples can be found in my bookFooled by Technical Analysis: The perils of charting, backtesting and data-mining.

References

Papoulis, A. Probability, Random Variables, and Stochastic Processes by 1965. McGraw-Hill

Disclaimer