Backtest Depth - Long-term Robustness (10 years) vs. Recent Adaptation (5 years)? - page 3
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
20s with real tick data and 5y ? I want to see that.
It is 100% exact, no approximations at all.
Took 432 seconds to calculate the history from 2020/01/02 till today for 15 algos on my lowly desktop (Intel Core i5-7400 @ 3.00GHz, 32660 MB), 28.8 seconds per algo. And this is the slow trader version which runs MathLog twice on every tick considering it is life trading the non-custom symbol. I have 2 versions of the EA, 1 for trading with all the checks and balances and one specifically for tester version which runs on custom symbol and uses EAtoMAth, then the caclulation time for 1 algo is little over 1 seconds for one pass over the same period, same symbol.
The calculated trade history of these 15 algos combined over the time window is over 600k: 550k VIRTUAL_2 the unfiltered history and over 50k VIRTUAL_1 the filtered hisotry.
And my code is still slow. I once shared a screenshot with someone orders of magnitues smarter and more capable then me where the tick count and calculated times where visible, i do not remember exactly the context, all i remembered him commenting my slow code and the need to optimize.
Fast code == more research in less time. Time is money you will never get back.
It is 100% exact, no approximations at all.
Took 432 seconds to calculate the history from 2020/01/02 till today for 15 algos on my lowly desktop (Intel Core i5-7400 @ 3.00GHz, 32660 MB), 28.8 seconds per algo. And this is the slow trader version which runs MathLog twice on every tick considering it is life trading the non-custom symbol. I have 2 versions of the EA, 1 for trading with all the checks and balances and one specifically for tester version which runs on custom symbol and uses EAtoMAth, then the caclulation time for 1 algo is little over 1 seconds for one pass over the same period, same symbol.
The calculated trade history of these 15 algos combined over the time window is over 600k: 550k VIRTUAL_2 the unfiltered history and over 50k VIRTUAL_1 the filtered hisotry.
And my code is still slow. I once shared a screenshot with someone orders of magnitues smarter and more capable then me where the tick count and calculated times where visible, i do not remember exactly the context, all i remembered him commenting my slow code and the need to optimize.
Fast code == more research in less time. Time is money you will never get back.
I got it, it reads the ticks from a file. But what about when it comes to trade simulation ? With the real ticks test there's a already biais, do you have live results that match your tests with eatomath ?
There is no difference in normal real ticks mode, virtual or EAToMath. All 100% identical down the mcs.
In normal real ticks mode, there is no bias. Explain what bias. I trade 2 strategies, breakout with market orders, mean reversion with limit orders. The difference in live vs backtest (or virtual/EAtoMath) is slippage in both cases, or partial fills or no fills with limit orders.
I use the same tickfeed for testing and live trading. Slippage and partial fill can be different because of the way the tester works. In tester there is no partial fill, and slippage is the result of the next tick (or artificial delay parameter) in tester whereas there is no next tick that affects slippage in live mode. Apart from that, the results in tester and live are identical.
There is no difference in normal real ticks mode, virtual or EAToMath. All 100% identical down the mcs.
In normal real ticks mode, there is no bias. Explain what bias. I trade 2 strategies, breakout with market orders, mean reversion with limit orders. The difference in live vs backtest (or virtual/EAtoMath) is slippage in both cases, or partial fills or no fills with limit orders.
I use the same tickfeed for testing and live trading. Slippage and partial fill can be different because of the way the tester works. In tester there is no partial fill, and slippage is the result of the next tick (or artificial delay parameter) in tester whereas there is no next tick that affects slippage in live mode. Apart from that, the results in tester and live are identical.
Damn. So you are basically re-creating the whole tester "framework" around that tick feed? If so, how do you efficiently handle complex event accounting (like partial closes, swaps, or grid/hedging positions) inside your custom loop without performance bottlenecks ?
Because that is exactly what the tester is doing, and that is what actually consumes so much memory. The data reading itself isn't what makes it so resource-heavy.
There are python frameworks (i forgot the name) that do this as well. But even they, as soon as it comes to virtually managing positions, they take their sweet time.Damn. So you are basically re-creating the whole tester "framework" around that tick feed? If so, how do you efficiently handle complex event accounting (like partial closes, swaps, or grid/hedging positions) inside your custom loop without performance bottlenecks ?
Because that is exactly what the tester is doing, and that is what actually consumes so much memory. The data reading itself isn't what makes it so resource-heavy.
I re-create nothing. The virtual library is basically a tester in itself with a few nuances. Just like the normal tester in profit in pips mode takes away curency conversions making it faster, Math mode is the most bare bone basic mode, there are no margin checks, nothing, therefor the fastest. You just need EAtoMath in order to use it for trading purposes as there are no ticks in math mode. It literally enables you to use it like real ticks mode only exponentially faster.
There is no custom loop. Everything is the same as normal.
The main thing is, if your 100 order grid is computed very inefficently, things will be faster with math mode, but still slow compared to not open orders 1000 pisp away for example. It all comes down to optimize uyour code. The reason why i work on a slow, deat beat desktop is, when i created something slow, i inmidately notice and try to optimize, when using fast hardware coders become lazy and do not care about optimization because it is more difficult to notice.
Optimize your code!
I re-create nothing. The virtual library is basically a tester in itself with a few nuances. Just like the normal tester in profit in pips mode takes away curency conversions making it faster, Math mode is the most bare bone basic mode, there are no margin checks, nothing, therefor the fastest. You just need EAtoMath in order to use it for trading purposes as there are no ticks in math mode. It literally enables you to use it like real ticks mode only exponentially faster.
There is no custom loop. Everything is the same as normal.