Reversion to the mean trading is a statistical form of trading based on the idea that while the price will fluctuate between highs and lows it can generally be relied upon to return to it's mean (or average) value with enough certainty to make trading the concept viable. To put it simply, mean reversion trading is waiting for a stock, commodity, currency or whatever to deviate significantly from it's 'true' value, that is, it's average or mean value, and then bet on the price returning back to it's true (average/mean) value.
To many investors and speculators this reversion to the mean trading strategy sounds viable and is very attractive, however I can see at least four major flaws in the concept although I'm sure that there are more. They are –
One. Markets often trend. If a market is moving in a strong uptrend and is above it's average price and a trader bets on it falling back to it's average value it may very well continue to move up and never return back down to the average. Even when the uptrend ends it may not fall back that far and even if it does it may take many years for this fall to occur which is far to long to make trades like this worth while.
Two. Market prices aren't normally distributed, while it often appears that market prices are normally distributed every now and again, say every 5 or 10 years, movements happen that are so large and significant that we should not expect to see them if prices were normally distributed. In short, financial markets are far more vulnerable to 'black swan' events than the Gaussian function or bell curve probability theory says that they should be. Yet it is on this probability theory that many of those trading systems are actually based. The story of Long Term Capital Management is a prefect example of trading using systems and strategies that don't properly account for the probability of black swan events.
Three. Occasionally there is a great difference between what is the average (mean) price and the median (middle) price. For example, take the following sequence of numbers 4, 5, 10, 60, 75. The middle value in the sequence is 10, but the average is wildly different, it is 30.8. However if a price moved between a low of 4 and a high of 75 the middle price in the range would be 39.5. In short, outlying prices can seriously skew the calculation of the mean value.
And last but not least, as the old saying goes: if it were really that simple, everyone would be doing it. If market prices really were normally distributed and the certainty of a reversion back to the mean could always be calculated mathematically everyone would be doing it and the strategy would lose it's edge.
Using Reversion To The Mean as a trading concept,
Despite all the flaws I have mentioned, I am not convinced that reversion to the mean as a concept is entirely without merit. Markets often do deviate and then correct themselves and all markets go through long periods of time when they are not strongly trending.
I am therefore going to try and make use of the reversion to the mean trading strategy by making an end of week trading system where we target a certain price in the following week based on middle values of the previous two weeks. When the market closes on a Friday I will take the week's middle value between the weekly high and low and compare that value to the middle value between the previous weeks high and low. At the end of the trading week the following week's target will be: (the middle value of the week just ended's range – the previous weeks middle value of it's range) + the middle value between the high and low of the week just ended. Or, in pseudo code: Target = (Middle_Value - Previous_Middle_Value) + Middle_Value.
This trading strategy could be considered to be a simplified and mechanical method of trading using Andrew's Pitchfork. I will test this system on the four major currency pairs (the EUR/USD, GBP/USD, USD/JPY and USD/CHF) and on four major stock market indices (the S&P 500, the FTSE, the German DAX and the French CAC) from the beginning of the year 2000 to the middle of 2011. The results are as follows -
Reversion To The Mean: The Results,
Market
Number of Winning Trades
Number of Losing Trades
Percentage Of Winning Trades
Number Of Points Won
Number Of Points Lost
Win To Loss Ratio (Points)
EUR/USD
417
175
70.44%
28175.0
27842.0
1.012 to 1
GBP/USD
406
187
66.95%
32746.5
35473.0
0.923 to 1
USD/JPY
429
174
71.14%
24003.0
22438.0
1.070 to 1
USD/CHF
430
174
71.19%
32533.5
30206.0
1.077 to 1
S&P 500
398
206
65.89%
4492.55
5109.95
0.879 to 1
FTSE
418
187
69.09%
20721.50
20987.00
0.987 to 1
DAX
389
219
64.30%
23809.06
31296.05
0.761 to 1
CAC
403
202
66.61%
21027.71
21313.00
0.987 to 1
When I first looked at the results of these tests I was some what surprised. I had expected to find an edge in projecting the median based on the previous two weekly candle's median prices and then targeting that middle value. But alas I did not find it. I did not find an edge in targeting a projected median value. That is not to say that no edge exists, but rather than I have no evidence of one and failed to find it. I will perform more tests and look into this further in the future.
