Backtest results

General
1

Hi,

I have a general question;
Is it normal to have different backtest results with the same strategy but in different testing timeframes (example 1min vs. 15min)
I used ATR Indicator as TP and the resultsarevery strange for me (low timeframe very bad results and high timeframe very good results)

Thanks

2

Lower the timeframe better the result of backtest.

3

No its just the oposit… when I test in 1min I have very bad results when I test in 1h I have very good results…in my opinion t should be the same result no matter what timeframe I test. As of my understanding its just the download of date which is much bigger in lower timeframes.

4

Backtester works with candles, not trades. We assume specific direction of price movement inside candle. The lower timeframe the more accurate results

5

okay understand, so if I want to have a very accurate result I need to test in the 1min timeframe correct?

6

Yes, correct

7

Told you

8

What would you expect if we would compare the results of

  • RSI-7 (5min) versus
  • RSI-35 (1min)?
9

@d_yo_r are you asking me or someone else?

1 Like
10

This is a general question. I haven’t tested it. I would expect the results to be similar or identical because RSI uses candles of the same 35-min timeframes even though with different precisions.

11

If the strategy uses indicators between 15m and 4h timeframe for example I wouldn’t expect any differences in terms of only the entries (or exits) between backtesting 15m and 1m timeframes assuming the same period was used.

What actually causes the changes are the intra candle events: stop losses hit, safety orders getting filled, market vs limit orders, slippage, etc

All these events are very time sensitive when it comes to the backtesting results.

And in particular if you place orders very narrow to the price or to each other in the case of SOs.

In extreme cases not even the 1min backtest can figure the order of the events and some assumptions must be made which are not always done correctly.

And this is not an issue with gainium, you will face it on any backtesting tool.

Solution? Real order book backtesting. Unfortunately the trade off is huge and not worth it since it involves a high use of resources and extremely slow as well.

This solution is also only a partial one since it doesn’t simulate the conditions of real time trading on an actual exchange either.

Still with all the limitations backtesting is critical to be able to distinguish between a promising strategy and a waste of time, it’s only the begging of a long journey, but at least one worth taking

2 Likes
12

Thanks for your answer!

Yes, any test mainly gives us an idea of what possible results might look like because it’s a simulation. Backtests are the first attempt to give us a hypothesis based on historical candles’ data, which lacks interbar data and only runs on OHLC. Paper trading has interbar data for the current candle, but can only be run in real time. Both lack the simulation of delays due to low liquidity or exchanges’ limitations and the resulting slippage.

Within these limits the results should be somewhat consistent if the same time frames are used. Even if the resolution of the candles may be different, and if no step uses random data, e.g. the Random Pair Selection.

1 Like
13

Sorry for this off_topic response first. Did you read my comment with calculations at one of your videos where you suggest a strategy to save safety orders?

14

Which video is it?

15

I created a separate topic for it.