Meta-Trader - Fx-Regression Live vs Backtest Results

Welcome back Meta-Traders.

Here are the results of a back to live test consistency check on my own Expert Advisor called FX-Regression. Since the system back-tests fairly well, the goal is to determine if these back tests are somehow flawed and will not be reflected in actual trading.

For the period of January 12, 2011 (when I started trading live) through March 31, 2011, here are the results:

- Results of trades was plus $41.71 or about +4.7

- Total Net due to Interest roll was minus $3.19

- Net P&L was plus $38.52 or about +3.5%

The period does not include the trade opened on 3/31 and 4/1 which resulted in a loss and put the account back into the negative. Attempts to reconcile the actual and live trades ran into these challenges:

- Back tests run on Forex.com showed a high number of chart mismatch errors, so the results had to be discarded

- Back tests were derived from Alpari US instead

- We had a clock change due to daylight savings. For live trading, I changed the start time for the trade to keep the start time consistent with US Eastern time, caused a shift in start time on the server that was not properly reflected in the back test.

Overall, the back tests showed the account down $60 or minus 6% for the test period. A detailed comparison of the trades for each symbol showed the following:

- EUR/USD took 40 live trades an about 30 of them were consistent the back test

- USD/CHF took 29 live trades and only 12 were consistent with the back test

- USD/CAD took 24 live trades and only 9 were constent with the back test

For cases where the live trading and back test didn't check out, we saw nearly every possible situation:

- Profits in live trading that showed as losses in the back test

- Profits in live trading that were missing altogether from the back test

- Losses in live trading showing as profits in the back test

- Losses in live trading that were missing altogether from the back test

Conclusions are difficult to come by with such a jumble of data. The fact that I got the closest results with EUR/USD show the spread-senstive nature of this trading system. Back tests used a fixed spread of 2 pips for EUR/USD, 3.5 for USD/CHF and 5.0 for USD/CAD set with the Asirikuy back test setting program. Differences between these values would cause the exact symptoms were are seeing:

- Back testing missing trades that executed in live trading

- Live trading taking profits that turn into losses in back-testing

But it doesn't necessarily account for cases where the back-testing was profitalble but live trading was not. These may be accounted for by differences between the simulated feed used for back testing, and the real-time feed used for trading

Overall conclusion is that back tests with this system bear at best a sketchy resemblance with live test results. Larger SP and TL values would reduce the size of the error cases not by eliminating the underlying problems - but by reducing the number of times they occur. Actual trading results - at least for the test period - were better that back test results. Whether this conclusion pans out for the rest of the year remains to be seen.

Have a great week all.