Fibonacci numbers are much used in both forex and futures trading. Honestly if I had a Dollar for every time somebody has said words along the lines of “and it got nearly right on the n% retracement – it was amazing” .
So it was interesting to read some research that has recently been published on whether Fibonacci levels can be scientifically proven to be hit more frequently than would be expected by pure chance.
The research was carried out by the academic Roy Batchelor, HSBC Professor of Banking and Finance at Sir John Cass Business School and PhD researcher Richard Ramyar.
They make the very true point that when trying to test the validity or otherwise of such concepts the problem is often that traders fail to specify their trading rules precisely enough to be scientifically tested. Traders tend to report outcomes using phrases like “price nearly got to” “price went just slightly past …then” etc. Such descriptions of trading results and outcomes is far too subjective to satisfy rigorous external verification.
To cut a long story short our trusty scientists tested daily Dow data from 1915 through to 2003 to see if after a long trend retraces is the magnitude of the next trend entirely random or does it respect support and resistance levels dictated by Fibonacci ratios or round numbers.
It often does seem that price does respect (or at least nearly respect) the predictions from Fibonacci levels and many other indicators such as RSI or MACD etc. Certainly when you simply “eyeball” the charts such correlations look very real … but what did the scientific testing tell us ?
Their conclusion after a great deal of testing was ….. “Our conclusion must be that there is no significant difference between the frequencies with which price and time ratios occur in cycles in the Dow Jones Industrial Average, and frequencies which we would expect to occur at random in such a time series”
So the answer is No folks (at laest not for daily data on the Down over that test period) – any correlations are no better than random – sorry about that.
This does not surprise me greatly as I have done a great deal of computer testing on all sorts of indicators and that would be my conclusion as well (on virtually all indicators not just Fibs) . Sure sometimes indicator A gives a great buy signal but then again sometimes it does not !
Although please note that whilst something may not “work” in the precise scientific sense that does not mean that it may not still be useful to traders in a more general sense. See the later article “why are people convinced Fibonacci works”
The great thing about algorithmic trading is that we can do our own testing and do not need to rely on the good Professor.
The litmus test for Algo traders is very simple “can I design mathematical trading system using Fibonacci on its own that will generate profits over a statistically valid period of trades and time”.
I have never managed to that. More importantly I am not aware of any mechanical mathematical trading systems that have managed to do so either. That answers the Fibonacci question well enough for me because as algorithmic traders we are only interested in trading methods that can be proven to work and generate profits. I will leave all the “nearly” and “within a whisker of” to others. I want stuff that absolutely stone cold can be proven to have worked.
This is one of the big advantages of the automated trading course it teaches you how to test your ideas in a similarly rigorous scientific way to enable you to reach your firm clear unequivocal conclusions rather than never really knowing if the method or indicator you are using really works or not.