Backtesting - worthless?
Hi,
I had the same problem and the data from metaquotes is terrible. If you are testing on Forex pairs, you can download the 1Min candle data from dukascopy historical feed and use the PeriodConverter script to create higher timeframes.
I know this from experience that Metaquotes data has some missing candles from time to time.
I got data from dukascopy and verified the backtest by comparing to trades on live Demo account.
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Hi everyone,
I'm interested to see people's appoaches when backtesting automated strategies...
Here's my journey with backtesting so far:
1. I initially started down the route of using MT4 optimisation and strategy tester to find 'profitable' EAs which I have written. I then realised the hard way the negative effects of slippage, swaps, comms and spreads, as what would appear to be a profitable strategy during backtesting very rarely was.
2. Having gone round in circles in the 1st stage for months, I then worked on improving the realism of my backtesting by testing on very wide spreads, calculating comms charges etc. Whilst this lowered the performance of the backtests, when testing on a live account and then backtesting over the same period, the results were astonishingly different.
3. I then learnt how 'bad' metaquotes data is and how useless it is for backtesting, and began to use TickDataSuite to backtest with tick data from multiple sources, factoring in slippage, comms and variable spreads. This worked to some extent as the strategy tester was now showing 'profitable' EAs as unprofitable, however it took a week to backtest 1 pair over a 3 year period - meaning it would take me 28 weeks to test every pair...
My current method is using LIVE market condition demo accounts from our broker (serious strings had to be pulled to aquire these, including a large 6 figure deposit). I have 4 EAs which are logged into these accounts hooked up to VPSs, testing EVERY currency pair to simple harvest live trading data.
At the end of each trading week, I then run a backtest of that week using TDS data and compare the discrepencies. There still seem to be alot, but the overall equity curves seem to be largely correlated.
I have been doing this method for around 6 weeks now and have aquired data for between 500-1000 trades on each algo - showing some pairs have been very profitable.
My question now is what do I do with this data???
Do I:
Option 1: Consider data from 1000+ trades to be more valuable than any data which could be harvested from backtesting. Start to trade the algo on the profitable pairs, but keep the 28-pair 'demo' account running to monitor all pair performance, allowing me to swap pairs in and out weeky.
Option 2: Consider that whilst this is a good data set, it is only from over a month of trading and so the data shoudn't be relied on yet... ie. keep testing.
Option 3: Other suggestions?
I appreciate that people don't like to share 'secrets' that they have discovered to help make profitable systems, however I think it is clear that I have invested ALOT of time, money and effort into the development process of EAs (to the point when I now understand that writing the code is the most simple bit!) so advice would be massively welcome and appreciated.
Thankyou.