Would You Trade This System?

The large number of trades in backtesting as well as its time-span is very good.

- I would look at all the stats on a year-per-year basis ... how much variability do you get?

- I would manually verify every single trade (in the backtest) for at least 1 year ... I know this is a lot of work, but the only way to ensure your strategy does what it is supposed to do.

- Re. the leverage discussion, I would trade this system on ES instead of SPY ... BTW, does the backtesting on ES gives similar performance results? Do you get mostly the same trades on ES & SPY? If not, do you understand why?

- How will you know (trading the system live) if/when this strategy has lost its edge? (ie, what are your system stops) - using these criterias, did the system lose its edge during the 14 years of backtesting?

- If I am not mistaken, your largest loss was about $2.8 / share ... that certainly seems heavy for an intraday strategy. What is you largest win, in comparison?
 
Quote from ssrrkk:

One thing that puzzles me is your PL curve hardly shows any draw-downs even when your sharpe ratio is low, and your win rate is in the low 60s, especially when your average losses are larger than your gains. I would have thought that with those stats, there would be more visible draw downs.
More...
Agreed.

The equity curve looks too smooth for a low Sharpe, while the other results sound unlikely to produce a high Sharpe.

The strategy appears to be pretty high risk per return, but that's not showing up on the equity curve. So .. something's probably not quite right.

Regarding backtesting being useful or not, my view is backtesting is excellent tool to use to rule things out.

If it's a dog in backtest, it probably will be when live.

If it's gold in backtest, then it *might* be ok when live.

A lot of the time, the market will react to you when live. Not so much in a backtest.

Honestly, I use backtesting mostly to debug execution algorithms, to expose incorrect assumptions, and to isolate the least-worst ideas.
 
Quote from ssrrkk:

One thing that puzzles me is your PL curve hardly shows any draw-downs even when your sharpe ratio is low, and your win rate is in the low 60s, especially when your average losses are larger than your gains. I would have thought that with those stats, there would be more visible draw downs.
More...


This is because there are so many trades fitted into that curve, let me post last years curve alone from same settings to give you more idea. Because yes there are dradwown periods and periods of making no money in there, its just over that huge of a period you do not notice them. But also the frequency of trades helps get out of drawdowns quickly. please look at the original post on first page to see the winning months and losing months. You will see there is losing months and its not a straight line up.
 

Attachments

spy report from last 12 months, see how individual years profit factors differ?

Before you only see the whole 10 year curve, if i break things down into individual years it gives more insight.
 

Attachments

Quote from tenthousandmen:

What is your sharpe? If it is below 1, it really isn't very good. You should be aiming for 2 or higher on the sharpe. 1.4 or above will do.

If this is scalping, you should also be concerned to how many slippage ticks before it becomes unprofitable. If it's 2 or 3, you have a problem when in bad market conditions.

You've probably made sure it's not false curve fits.

These are my thoughts. Cheers..
More...



Well im not scalping a couple of ticks. Ave is 20 cents id say but it changes with volatility. The slippage is in because all entry and exit fills must exceed price before being counted if you know what i mean.

Sharpe ratio is low agreed.


thanks for your feedback, appreciate it.
 
Quote from dom993:

The large number of trades in backtesting as well as its time-span is very good.

- I would look at all the stats on a year-per-year basis ... how much variability do you get?

- I would manually verify every single trade (in the backtest) for at least 1 year ... I know this is a lot of work, but the only way to ensure your strategy does what it is supposed to do.

- Re. the leverage discussion, I would trade this system on ES instead of SPY ... BTW, does the backtesting on ES gives similar performance results? Do you get mostly the same trades on ES & SPY? If not, do you understand why?

- How will you know (trading the system live) if/when this strategy has lost its edge? (ie, what are your system stops) - using these criterias, did the system lose its edge during the 14 years of backtesting?

- If I am not mistaken, your largest loss was about $2.8 / share ... that certainly seems heavy for an intraday strategy. What is you largest win, in comparison?
More...


I have already verified every trade and yes code does exactly what it supposed to......

There was a couple of bad years in there but still managed to grind out a profit.

Yes some wins are less than the losses, and in volatile times the losses can be large. But statistically over the sample you see the lossess are well contained and within risk models i feel. I have used the stats to form probabilities to which stop and target levels are the most suited over the long term.

A rr of less than 1-1 is not ideal, but when i increase to 3-1, 2-1 the performance and choppiness gets worse but still net profitable every year. This must mean my entry is not ideal and could be improved i guess, But i personally do not like to focus on perfect entry because this drives me mad due to the randomness of the markets. But saying this, it could be that SPY, Es and main market instruments are more ranging then trending so capturing smaller targets is easier?............i do need to capture some trends though in order to increase profit factor so maybe stocks are the answer and ones that are trendable

Alot of pro's use 1-1 or less believe it or not. They counter act this by high volume, high frequency trading. You could not get away with this approach just trading 1 trade a day or something like that.

This is just one system. The idea is to be like lescor and have 5/6 similar non correlated ones to smooth further. Ill post up a couple more systems i have to get feedback if possible
 
before i post up some more systems, i would first like to show the worst year for the system which was 2005 if you see from the stats on the first page.

Please take a look to show a bad year to give more insight.......



Please see that this is not a perfect system and a very average one. Look at the worst year here and notice how for 200 trades it was in drawdown. Thats half a year!, not ideal.....
 

Attachments

yes ES is very similar performance as they are both more or less the same instrument.

I just use spy for some reason. I think because im put off futures from bad experiences in the past. Also the plan was to take this to a basket of stocks that it worked on. Which is still work in action right now.
 
It has not been fitted to a specific time period.....For example if i post performance starting 6 months ago, it still gives similar performance.

I suppose though if i did start the test at differing time periods i would eventually get a drawdown year but only one i feel. I would have to fit though to the losing year.


Look at the monthly net profit graph on page 3 to see what i mean. And where worst drawdown period comes.
 
Quote from jayjay121:

This is because there are so many trades fitted into that curve, let me post last years curve alone from same settings to give you more idea. Because yes there are dradwown periods and periods of making no money in there, its just over that huge of a period you do not notice them. But also the frequency of trades helps get out of drawdowns quickly. please look at the original post on first page to see the winning months and losing months. You will see there is losing months and its not a straight line up.
More...

If this is a 1-min timeframe, prifits from trades should be almost normally distributed. But you start making profit after the 150 trade or so for this year.

This does not make any sense unless you are using several filters with optimized parameters.

Are you using work-forward optimization in your testing?

Your results are highly unusual.

1) Over-fitting far and beyond

or

2) Look ahead in backtest

or

3) Unrealistic settings
 
You must log in or register to reply here.
Back
Top