Need comment before going live...

[ShadowKnows]

Hi,

Please see the attached picture about my strategy result. It is tested on ES with 3-years data & 10 contracts per entry.

I am new to NinjaTrader (so as program trading), I need your advise if you may have been around with live experiences and seen many strategy results...

(1) Does it look like a tradable strategy to go live? Is it good or bad?

(2) Any data looks suspicious I should be aware of and do further adjustments upon?

(3) Any further testing do you suggest before going live?


Thanks a lot for your comment & help.

[Baruch]

I would test it with atleast 20 contracts.

[NinjaTrader_CalH]

Hi ShadowKnows,

Thanks for posting.

Although I can't offer you advice on if the testing looks good to go live or not, I can give you a link on the real-time vs. backtest and what to expect.
http://www.ninjatrader.com/support/h...ime_vs_bac.htm

With real-time vs. backtest, you should expect that a strategy running real-time will produce different results than the performance results generated during a backtest.

You may have different results running your strategy in real-time than you do in backtesting. For this reason you can test this strategy with the SIM account in a live feed/simulated data feed or use Market Replay Data to test out the strategy.

Let me know if I can be of further assistance

[koganam]

Hi,

Please see the attached picture about my strategy result. It is tested on ES with 3-years data & 10 contracts per entry.

I am new to NinjaTrader (so as program trading), I need your advise if you may have been around with live experiences and seen many strategy results...

(1) Does it look like a tradable strategy to go live? Is it good or bad?

(2) Any data looks suspicious I should be aware of and do further adjustments upon?

(3) Any further testing do you suggest before going live?


Thanks a lot for your comment & help.

Be careful. Your results seem to imply that you are using a StopLoss about twice your profit target. If that is the case, a neutral edge should give you a success trade rate of 66%, so you appear to have an edge.

1. My question would be that with an edge, do you really want to be trading with a risk 2 times your reward?
2. How does the system react to drawdowns?
3. What happens when you take commissions into account on 10 contracts per shot?
4. With a realistic account size, would the largest drawdown have put you into a margin call?

In other words, while your summary looks nice, you want to take a look at your equity curve. I could probably give more comments if I saw the equity curve and "Profit Per Month" graphs. Are your results because of a few large successful trades, or because you steadily make more than you lose?

[ShadowKnows]

Be careful. Your results seem to imply that you are using a StopLoss about twice your profit target. If that is the case, a neutral edge should give you a success trade rate of 66%, so you appear to have an edge.

1. My question would be that with an edge, do you rally want to be trading with a risk 2 times your reward?
2. How does the system react to drawdowns?
3. What happens when you take commissions into account on 10 contracts per shot?
4. With a realistic accoutn size, would the largest drawdown have put you into a margin call?

In other words, while your summary looks nice, you want to take a look at your equity curve. I cold probably give more comments if I saw the equity curve and "Profit Per Month" graphs. Are you results because of a few large successful trades, or because you steadily make more than you lose?

Thanks for pointing out the profit & stop ratio.

(1) It is currently the best case, I will try to see if I can improve it.

(2)(4) Please see attached graph & period report about the drawdowns.

(3) There are 565 trades and that is 5650 contracts. The commission should be under $20000 in this case.

Thanks for your help.

[ShadowKnows]

I would test it with atleast 20 contracts.

Without going live, I think 20 contracts just mean the figures double in the summary report. Isn't it?

[sledge]

Without going live, I think 20 contracts just mean the figures double in the summary report. Isn't it?

You should really market replay your item first.

I do know that multiple fills at different levels result in multiple wins or losses. You need to force NT to fill all to get realistic win %, else it is skewed.

Say you have enter long 5 contracts, 3 fill at xx.xx, 2 fill at xx.yy.. that's 2 wins!!!


Anyways - to answer your question about doubling - it should hold at 20 contracts, although you might start seeing more different fill prices showing up.

[koganam]

Thanks for pointing out the profit & stop ratio.

(1) It is currently the best case, I will try to see if I can improve it.

(2)(4) Please see attached graph & period report about the drawdowns.

(3) There are 565 trades and that is 5650 contracts. The commission should be under $20000 in this case.

Thanks for your help.

Interesting results. Now you really want to include commissions, and test with only one contract, so that you know your returns per contract, in currency terms. That is because, once you do so, some of those marginal winning months will correctly show as small losses.

Then you will want to put it through Market Replay. I do not know the details of your strategy, but if it uses anything other than fixed-period or Range bars, it is practically mandatory to do the Market Replay. If you are using Renko bars, it IS mandatory, or you will have a very wrong impression of what is going on.

[ShadowKnows]

Interesting results. Now you really want to include commissions, and test with only one contract, so that you know your returns per contract, in currency terms. That is because, once you do so, some of those marginal winning months will correctly show as small losses.

Then you will want to put it through Market Replay. I do not know the details of your strategy, but if it uses anything other than fixed-period or Range bars, it is practically mandatory to do the Market Replay. If you are using Renko bars, it IS mandatory, or you will have a very wrong impression of what is going on.

Yes, I understand the difference in simulation & live trading...

but in general, does it look like a good system to trade on from the simulation results?

[Baruch]

Without going live, I think 20 contracts just mean the figures double in the summary report. Isn't it?

It was a joke. You have to read it - test with only ONE contract.

[koganam]

Yes, I understand the difference in simulation & live trading...

but in general, does it look like a good system to trade on from the simulation results?

1. On what time frame and on what kind of bars was the strategy tested?
2. Have you run a Walk-Forward test to detemine how the strategy behaves when it does not have perfect knowledge of the future, which in effect, is what a full period optimization represents?

[ShadowKnows]

1. On what time frame and on what kind of bars was the strategy tested?
2. Have you run a Walk-Forward test to detemine how the strategy behaves when it does not have perfect knowledge of the future, which in effect, is what a full period optimization represents?

1. it's 3 mins timeframe with normal candle bars

2. my strategy has many parameters, and I don't really understand Walk Forward. Whenever I use it, it seems running forever.
Instead, I optimized the parameters for one year, and then use the optimized parameters to test another year and another year. During each optimization on each year, I adjusted the parameters in between, and finally come up with the optimized parameters for three years combined.

Now I even have more optimized parameters. Here is the graph.

[koganam]

1. it's 3 mins timeframe with normal candle bars

2. my strategy has many parameters, and I don't really understand Walk Forward. Whenever I use it, it seems running forever.
Instead, I optimized the parameters for one year, and then use the optimized parameters to test another year and another year. During each optimization on each year, I adjusted the parameters in between, and finally come up with the optimized parameters for three years combined.

Now I even have more optimized parameters. Here is the graph.

Given what you now describe, you appear to be saying that you performed a semi-manual WFO, using an optimization period of 1 year and a test period of 2 years. It would appear then that the only things left are to enable commissions and run a Market Replay.

Just to be realistic, however, notice that in the very first month, your strategy lost money. On the assumption that you do not have perfect knowledge of the future (which is what curve-fitting optimization gives you), would you really have traded this strategy if it lost money in the very first month? That is the problem with optimizing, and it could very well point to curve fitting.

If you choose to go forward with this, you might think of starting with one contract, and rev up as you make money. Starting off with 10 contracts on a system that has not yet proven itself in live markets strikes me as too chancy.

[ShadowKnows]

Given what you now describe, you appear to be saying that you performed a semi-manual WFO, using an optimization period of 1 year and a test period of 2 years. It would appear then that the only things left are to enable commissions and run a Market Replay.

Just to be realistic, however, notice that in the very first month, your strategy lost money. On the assumption that you do not have perfect knowledge of the future (which is what curve-fitting optimization gives you), would you really have traded this strategy if it lost money in the very first month? That is the problem with optimizing, and it could very well point to curve fitting.

If you choose to go forward with this, you might think of starting with one contract, and rev up as you make money. Starting off with 10 contracts on a system that has not yet proven itself in live markets strikes me as too chancy.

Thanks. Sure, I will start with 1 contract to test it first.

I have not seen other strategy results live, so I am wondering if this one looks good enough? (even if it might be curve-fitting optimization) Or I should aim for better results before considering going live?

[koganam]

Thanks. Sure, I will start with 1 contract to test it first.

I have not seen other strategy results live, so I am wondering if this one looks good enough? (even if it might be curve-fitting optimization) Or I should aim for better results before considering going live?

In the abstract, that is a nice equity curve. Only you can determine if you are comfortable with it.

Just my $0.02, I would never trade it, just because I personally would never trade a system that has parameters where the potential loss is so much larger than the potential win. This is because signal processing concepts tell me that such a system is unstable. Here is a post/thread which expands on the mathematics of this issue. ref: http://www.ninjatrader.com/support/f...53910#poststop. You might want to backtrack and read the whole thread.

YMMV.