Let’s discuss the difference of two expert advisors (fully automated) running in parallel at the same time (same currency pair, same broker, etc), one on a demo and the other on a live account. “Forward testing on a small live account is better than on a demo account.” Why? Let’s not discuss the psychological factor. We all know that it differs in: – slippage – requote – spread – commission costs – swap costs But what else does live differ from demo trading? – for example, is the data feed exactly the same or is there difference in manipulation?…
The only ‘real’ difference between demo trading and live trading, provided you are using a reputable broker is in slippage and sometimes exchange rate impacts and SWAP. For example in MT4 backtesting land, SWAP is not included in the results and typically the variable spread at the time of running the backtest is used as a proxy for the historical spread. The latter however can now be resolved in some backtest applications in the MT4 environment……however SWAP and Exchange rate impact on the base currency will only ever be based on using the current variables from your broker as a basis for historical application.
Slippage on the other hand is definitely a factor to also consider when moving from demo to live trading environments.
I think those that are looking for accuracy between Demo and live trading will always be disappointed because they will never agree. It is not usually through dodgy broker practices however….but rather the perturbation that exists in any price series of a liquid market.
Hopefully everyone who trades recognizes that it is probabilities that rule the game as opposed to preciseness. If you want to be precise when flipping a coin and determining if it will come up as heads and not tails…then be my guest…however the perturbation that exists in a single coin toss demonstrates that you can’t play the game this way. You need to apply the Law of Large numbers to come to a solution.
There is enough ‘noise’ in a return series from a live trade than there is in a demo one. You can see this when you plot where your ‘actual’ return series resides in a shuffling of that return series through bootstrapping.
If you get identical monkeys or even two identical seperate EA’s to live trade together on two separate accounts, this feature between the disparity between return streams from perturbations in data receipt or trade efficacy will become self-evident.
Each of those particular return series from the bootstrap chart above are entirely possible. This simply demonstrates that this game is not one of preciseness but probabilities. An edge will only play out over an extended data series just like a biased coin….so this therefore puts a different take on what you should be gaining from backtesting or forward testing your strategy.
Note that the bootstrap series above was generated from a strategy with only the faintest edge. In fact it was not a worthwhile strategy to progress no matter what the individual return stream produced looked like. This is one way to detect how ‘randomness’ can fool you. Only if the entire sequence of distributions displayed a positive return could you be satisfied that your strategy had enough cohonas to beat the noise of the market.
You are not testing for preciseness but rather the robustness of a general principle and whether that general principle holds up under the Law of Large numbers. The folks who get the most out of backtesting realise this principle is implicit in their results…..but are rather simply testing whether the general principle of say divergence or convergence play out over the long term and what scenarios could be generated by that deployed strategy over the long term and by applying the Law of Large numbers in your testing. It gives you a road map into uncertainty that allows you to determine whether your live strategy lies within the bounds of acceptability of that road map.
Quidquid latine dictum, altum videtur