Question regarding backtesting

[musslewhite]

I am running extensive stock backtesting with 15 minute candles, the situation is that the average life of my trade (from buy to sell) is only about 30 minutes (two candles)... The backtesting results are nothing near the real time paper-trading results.

Should I be using another type of candle, i.e. 1 minute candles?.

 

[HowardCohodas]

I am running extensive stock backtesting with 15 minute candles, the situation is that the average life of my trade (from buy to sell) is only about 30 minutes (two candles)... The backtesting results are nothing near the real time paper-trading results.

Should I be using another type of candle, i.e. 1 minute candles?.

Disappointment when moving from backtesting to paper trading is a frequent occurrence. There are limitations of backtesting that are important to understand before you convince yourself that it has potential to move to paper trading.

Before I elaborate, perhaps you will share what instrument you are trading.

 

[songcon]

I am running extensive stock backtesting with 15 minute candles, the situation is that the average life of my trade (from buy to sell) is only about 30 minutes (two candles)... The backtesting results are nothing near the real time paper-trading results.

Should I be using another type of candle, i.e. 1 minute candles?.

This is what I've learned from T.A.. If you see that there is a pattern
(cycle) is 30 minutes, you should use a 15 minute moving average for a
30 minute cycle. There is a technical term that is centering technique.

With that in mind, you should do additional research yourself on the subject.
What you are doing is correct - according to the book .

PS: From buy to sell is just 1/2 of the cycle: A cycle is defined as from low
to low. From the chart, you can determine the average length of the cycles.

This is more of a non-machine observation.

 

[Blaiserboy]

You did not mention as to whether you are doing manual or machine tests.

Certainly you should be looking at different time frames as that will help to establish your logic.

 

[meanreversion]

Disappointment when moving from backtesting to paper trading is a frequent occurrence.

Not for you though Howard, eh?

 

[HowardCohodas]

Not for you though Howard, eh?

Many, many, many times. However, persistence can sometimes lead to success. Especially when you finally understand the limitations of your approach.

Kind of like the lessons you should be learning.

 

[Rossini]

I am running extensive stock backtesting with 15 minute candles, the situation is that the average life of my trade (from buy to sell) is only about 30 minutes (two candles)... The backtesting results are nothing near the real time paper-trading results.

Should I be using another type of candle, i.e. 1 minute candles?.

Ignore the TF for now. Just focus on why back testing results are so different to live ones.
I have pinched the following from Adamus journal (I'm sure he wont mind):

Checklist of Stupid Mistakes giving Improbably Profitable Backtest Results

* check number of trades - not enough trades. e.g. 1 trade in whole period, always looks really profitable
* short test period with a big trend makes trades in one direction massively profitable
* exit at small target on same bar - always looks good
* forgotten about slippage and commission? That HFT system probably won't work unless you have Goldman Sachs' trading platform.
* if it's a backtest across a basket of instruments, one massively profitable instrument can make the whole basket look good - which reminds me: if one instrument in a basket is massively profitable, the weighting is probably out by a factor of 10.

Something to think about? Hope it helps.

 

[HowardCohodas]

Checklist of Stupid Mistakes giving Improbably Profitable Backtest Results

* check number of trades - not enough trades. e.g. 1 trade in whole period, always looks really profitable
* short test period with a big trend makes trades in one direction massively profitable
* exit at small target on same bar - always looks good
* forgotten about slippage and commission? That HFT system probably won't work unless you have Goldman Sachs' trading platform.
* if it's a backtest across a basket of instruments, one massively profitable instrument can make the whole basket look good - which reminds me: if one instrument in a basket is massively profitable, the weighting is probably out by a factor of 10.

Good ones all.

Perhaps one could add "not validating your results with out of sample data?"

 

[meanreversion]

Also, don't forget that the software has no knowledge of what happened DURING the bar.. all it knows is OHLC. Thus if your system buys at (e.g.) 100p with a profit target of 110p, then if the range of the bar is 90-120p, the system may assume that you entered at 100 and then sold at 110, even if the price opened at 120 and went down tick by tick to finish at 90.

It's a tough one to explain but if you do enough backtesting you'll understand it fairly quickly.

 

[Blaiserboy]

There are many aspects to backtesting, and there are many ways of examining results. I recently started to compile a list of things to look at.......

Backtesting can be an onerous job, but it is essential.

 

[Blaiserboy]

I am running extensive stock backtesting with 15 minute candles, the situation is that the average life of my trade (from buy to sell) is only about 30 minutes (two candles)... The backtesting results are nothing near the real time paper-trading results.

Should I be using another type of candle, i.e. 1 minute candles?.

You may have to look at OHLC in addition to candles

 

[Blaiserboy]

Some info on backtesting Candlesticks here

http://blog.kilotrader.com/2010_05_01_archive.html

It may not be applicable to your tests, but some interesting results