Why Do EAs Perform Differently in Backtesting and Live Trading?

New comment

I have been trading for a long time, and I have noticed that most EAs perform well in backtesting but poorly in live trading.

I even backtested an EA for the exact same month when it was running on a low-latency VPS. The backtest showed good results, but the actual trading performance was poor.

What could be the reason for such a big difference? Is it due to tick data, spreads, slippage, execution speed, or something else? How can we identify the real cause?

It was discussed many times during the last 10 years so I am suggesting to check this summary thread about:
All (not yet) about Strategy Tester, Optimization and Cloud 
and this post  as my experience for example.
All (not yet) about Strategy Tester, Optimization and Cloud
All (not yet) about Strategy Tester, Optimization and Cloud
  • 2013.08.20
  • www.mql5.com
The text provides an extensive overview of the MetaTrader 5 Strategy Tester, its functionalities, limitations, and usage scenarios. It covers testing strategies, optimization techniques, cloud network capabilities, and common issues faced by users. The content also discusses the differences between testing and optimization, the importance of real tick data, and the challenges of using the tester with custom data feeds.
If you run backtests using historical data on every real tick, with 100% historical data quality, there are only very minor differences. The MT5 tester is exceptionally good.
Ghenadie Tumco #:
If you run backtests using historical data on every real tick, with 100% historical data quality, there are only very minor differences. The MT5 tester is exceptionally good.
Confirming this.
Zeeshan Javaid:

I have been trading for a long time, and I have noticed that most EAs perform well in backtesting but poorly in live trading.

I even backtested an EA for the exact same month when it was running on a low-latency VPS. The backtest showed good results, but the actual trading performance was poor.

What could be the reason for such a big difference? Is it due to tick data, spreads, slippage, execution speed, or something else? How can we identify the real cause?

Hi Zeeshan,

testing the exact same month was already the right move. The tester itself is good, like Ghenadie says, so a big gap usually means the two runs weren't really the same.

I'd go one step further than Zufri and compare trade by trade, not only the totals. Export both lists for that month and put them next to each other by entry time.
If it's the same trades, just with worse prices live, its costs: spread, slippage, commission. That hurts, but it's normally small.
If there are trades that only show up in one of the two lists, that's the big one.

Then the EA didn't see the same thing:
other price data in the tester than on your broker, a different set file, or a filter that worked live but not in the test.
As far as i know the news calendar doesn't work in the MT5 tester, so a news filter is never part of the backtest.

The first trade that is missing on one side usually tells you more than any statistic.

Hope this helps :)
Chris
New comment