Low Quality Broker Data on MT5 - page 2
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
I used to do that in the past, like many people set out developing a strategy by backtesting various things.
But what i found was remarkable.
Most of the strategies that worked on a live feed, could not easily be backtested, think like multi currency strategies.
The strategies that were successful in the tester, rarely ever were on the live feed.
Other strategies that failed on the tester, were highly profitable on the live feed.
I still use the tester but only to check that the logic fires correctly.
If i need values, i run a mini backtest in the OnInit() function just a scan over the bars in history.
These are always up to date.
You can calculate crosovers really fast that way while backtesting to get these values, can be a time consuming task.
I see. It all comes down to preference I suppose.
I look at backtesting in a similar way to which I look at polling.
Polls are the best way for us to gauge the outcome of an upcoming election, but as was proved in 2016, they are by no means automatically indicative of the results in the actual election.
I think that backtesting is the best way for us to gauge the performance of a strategy, but as you mentioned, this is absolutely no guarantee that the results will be replicated in live trading.
As in the case of polls, I think backtesting is probably the only way for us to gauge the probability of a particular outcome.
I'm not sure, what exactly you mean mean by "most reliable method"!
You actually answered my question here:
On MT5 I will use the broker data for normal back-tests to evaluate the code functionality and fix bugs, or even for short crude optimisations runs, but when I want to do a serious analysis of how the EA will hold up and derive proper optimisation results, I will then use the Dukascopy data.
However, please be aware that if you use the MT5 Custom Symbol for testing, you will not be able to use the Cloud Testers.
But just to clarify, by 'best method', I mean testing using every tick based on real ticks in MT5 (with some missing data) vs. every tick method in MT4 (using external data).
... but when I want to do a serious analysis of how the EA will hold up and derive proper optimisation results, I will then use the Dukascopy data.
Okay so is it fair to assume that you value the quality of the historical data over the advent of native tick data?
I always use real tick data in tests, irrespective of using only the open price or not, because I use virtual stops in order to get real correct exit prices and not the fake invalid results that the Strategy Tester produces as it always sets the exit price at the exact price of the hard stop without any slippage what-so-ever.
When using reliable NND (non-dealing desk) brokers, the OHLC price data is almost identical because they often share the same feed and liquidity providers (there are only a few to go around), and the only thing I do is adjust the spread by padding the Dukascopy data to account fro spread differences between the brokers.
Can you explain to me at what threshold you believe the test becomes valuable, and in what respect it becomes valuable.
Also, in regards to the above, here are the results of a strategy which trades only at the open of each bar on the daily chart between 01/01/2008 - 01/01/2018, with three different brokers on the same currency pair.
Broker #1
Every tick based on real ticks
1-Minute OHLC
Net Profit
173.54
246.01
History Quality
60.00%
92.00%
Broker #2
Every tick based on real ticks
1-Minute OHLC
Net Profit
236.29
128.47
History Quality
60.00%
94.00%
Broker #3
Every tick based on real ticks
1-Minute OHLC
Net Profit
-306.22
-271.89
History Quality
60.00%
94.00%
All of the tests were conducted on EURGBP with GBP set as the test currency. Yet the results differ wildly just due to differing data from each broker. The results also vary by a fair amount depending of which testing method is used (1-minute OHLC vs. Every tick based on real ticks).
I always use real tick data in tests, irrespective of using only the open price or not, because I use virtual stops in order to get real correct exit prices and not the fake invalid results that the Strategy Tester produces as it always sets the exit price at the exact price of the hard stop without any slippage what-so-ever.
Okay thanks a lot for that input.
When using reliable NND (non-dealing desk) brokers, the OHLC price data is almost identical because they often share the same feed and liquidity providers (there are only a few to go around), and the only thing I do is adjust the spread by padding the Dukascopy data to account fro spread differences between the brokers.
So is the opposite of a non dealing desk broker a broker than uses an ECN? If not, then what do you define as a non dealing desk broker?
All my demo accounts are set up as standard accounts rather than ECN so this may explain the fact that my OHLC data IS different from one broker to another...?
Okay thanks a lot for that input.
So is the opposite of a non-dealing desk broker a broker than uses an ECN?
No! A true ECN broker is a NDD broker that uses Market Execution policy.
The Dealing-Desk brokers (the opposite of NDD) are the ones with Instant Execution policy and that take your position for you instead of placing it directly on the market (and usually trade against you).
PS! The Spread-Betting class of brokers in the UK are Dealing Desk brokers!
I think that backtesting is the best way for us to gauge the performance of a strategy, but as you mentioned, this is absolutely no guarantee that the results will be replicated in live trading.
This is what most people believe.
This usually applies to people that do not have a clue about what to do, and so they run 'trial and error' tests, where they change a parameter ever so slightly and re run the test again, to se if it outperforms previous tests.
But i can tell you something to think about.
It's not necessary to re-invent the wheel, it has all been done before...
When you take something, that has worked in the past, and that is still working till this very day,
The need to backtest anything, it simply disappears.
That is not trial and error but an exact mechanism that needs to be applied by exact means.
In fact i would encourage anybody from changing anything, out of fear that it just might stop working when you start to customize it.
That could save you a lot of backtesting, don't you think ?
The Dealing-Desk brokers (the opposite of NDD) are the ones with Instant Execution and that take your position for you instead of placing it directly on the market (and usually trade against you).
PS! The Spread-Betting class of brokers in the UK are Dealing Desk brokers!
Ah right, I'm with you now!
No! A true ECN broker is a NDD broker that uses Market Execution.
So just to confirm, testing using an ECN broker will give more reliable OHLC values/data than with a dealing desk broker?
When you take something, that has worked in the past, and that is still working till this very day,
The need to backtest anything, it simply disappears.
But how does one know in the first place if the system has worked in the past and still does to this very day if no backtesting has been undertaken?
That is not trial and error but an exact mechanism that needs to be applied by exact means.
This is one of backtesting's many flaws.
Ah right, I'm with you now!
So just to confirm, testing using an ECN broker will give more reliable OHLC values/data than with a dealing desk broker?
Again, you misunderstand. What I stated is that Dukascopy data is almost identical to the underlying data used by NDD brokers because they come from the same source. There are only a few data feed and liquidity providers so it is quite common for these brokers to be using the same source.