読み取り専用

フォーラムは現在、読み取り専用のアーカイブとなっています。.

バグ報告やプラットフォームに関する質問はこちら → [email protected]

私たちのコミュニティはDiscordとYouTubeで活動しています。ぜひ参加してください!

ご参加ください Discord YouTube

Help with Robustness Tests

2件の返信

BaobáTeam シモンズ

カスタマー、bpb_participant、コミュニティ、sq-ultimate、4件の回答。.

プロフィールを表示

1年前 #288953

皆さん、こんにちは,

I’m a beginner in the wonderful world of SQX and have some questions that I haven’t found conclusive answers to. I’d love to hear the opinions of more experienced users in the community.

I’ve been following all the available robustness tests in my pipeline. Overall, I find strategies that manage to pass the robustness tests (with specific adjustments for each asset and its nuances).

I’ve started noticing that strategies that did not pass Monte Carlo, WFM, or SPP also seem like good strategies. I ran a backtest on additional periods beyond those I used to build my strategy, and they showed positive results with similar statistics.

That said, I’d like to hear the opinions of more experienced members regarding robustness tests. I understand that passing all tests would make the statistics more reliable when deployed in a real environment, but how important is this for an algorithm’s success? Does it really make a difference when productizing an algorithm?

Since I have little experience with SQX, I don’t have enough data to refute the hypothesis that passing these tests leads to greater success in a real environment. I am testing it on a demo account and can share results later. 

Thanks in advance for your help and opinions!

1

ジェイソン

サブスクライバー、bbp_participant、カスタマー、コミュニティ、sq-ultimate、26件の返信。.

プロフィールを表示

1年前 #289002

I’ve been building algos for about 4 years now but I’ve only been using SQx for a few months. I bought the top tier license and I thought SQx’s training course was excellent. Over the years I’ve learned a few things. Heres what I think I know:
1. There is no golden test to determine if a strategy will perform on live data as it performed on historical, but a suite of testing gives you a better feel for how the strategy *Should* perform. That allows you to evaluate the strategy on live data with some confidence on what you can expect. This confidence on what you *can* expect should inform (provide feedback) on what criteria you personally want to set for the strat. For instance I don’t trade a strategy with less than 50% win rate. I find that allows me to evaluate a new strategy live quickly (20 – 30 trades) AND I’m more comfortable with that type of strategy. For instance if you traded a strategy with a win rate of 20%, you would necessarily have to allow for a LOT of losses to determine if you’re getting an acceptable equity curve.

2. The Last OOS test is very important to me. I’ve found if a strategy passes all 3 sections of historical data (i divide my historical data into 3 equal chunks) then it has a good chance of success on live data.

3. The Monte carlo tests are use for measuring certain aspects of your strategy. I particularly like the parameters monte carlo because it shows you the different possible ‘futures’ for your strategy. I like to see a strategy with a tight fan of different equity curves and I kick out strategies that have a weird equity curve. This part is a bit subjective but I’d rather fail a good strategy than pass a bad one. The CPU can look for new ones, it doesn’t mind.

4. The different market test and different timeframe tests give me a gut check on whether I’ve got a strategy that makes sense logically. My strategies don’t have to perform amazing on these test but they have to make money. If your strategy is successful across a wide range of different securities and timeframes the odds that its curve fit to one particularly security and timeframe are low.

After I’ve performed the Last OOS test (last 1/3 of data) I look for a smooth equity curve that I wouldn’t mind trading and it needs to make money consistently. To be honest, I don’t pass a strategy that lost money in the last couple years but if it has a losing year in like 2009 or something I consider passing it. Really depends on HOW much it lost in 1 year. I also like to have a look at the stagnation period. I don’t want a stagnation period great than 180 days. I try to model and grade strategies on what I would personally want to trade. I don’t think everyone does that but its how I do it.

If my strategy passes all that, then I run it through a WFA and I look at the permutations of the different parameters and I have a custom WFA setup that it needs to pass. I want to see that my PF and my drawdown are fairly close. Particularly the PF. I want to know that the most likely metrics are close to the backtested metrics. I used to think of the WFA as the holy grail of tests, but I don’t think that now. Some strategies probably don’t need a WFA, but I like to perform one to give my strategies an opportunity to adapt to changing market conditions.

The last part isn’t really a test, but I find it helpful. I don’t get excited about 1 strategy, I want a whole team of strategies on different securities that give me a nice smooth equity curve of uncorrelated returns. I would rather trade a whole bunch of profit factor 1.5 strategies on different futures products than a single PF 2.5 strategy. Lastly; don’t trade strategies that make you super nervous or you wouldn’t want to trade discretionary. For me that means I won’t trade a strategy with less than 50% win rate, I want a high ret/dd (8 or better, I’d prefer a 16 or better) and I don’t want a system that risks more than 2% of my account on a single trade. This is important because you need to have a reasonable amount of risk for the security you’re trading and the timeframe*ATR that you’re trading on. I think a lot of people trade mini’s not realizing a mini con on ES or NQ might need $2000 – $5000 per trade on the 1hr timeframe. If you’re risking $5k per trade and you follow the 2% rule, you need $250k in your trading account. Thats not a small amount of money. I follow the 2% rule and as a final gut check I take the backtest maximum drawdown and I double it. So if the maximum drawdown is $10k across the whole backtest, then I need to be comfortable with a $20k maximum loss (maybe comfort isn’t the word, but you get the idea). If the strat passes all those considerations, it goes into a hopper of tradeable strategies and I let SQx’s portfolio composer tell me what mix of strategies to trade.

Hope all that helps.

1

チャド・リン

定期購入者、bbp_participant、sq-ultimate、顧客、コミュニティ、39件の返信。.

プロフィールを表示

1年前 #289930

Thank you Jason, very impressive experiences!

Could you please share more experiences about SQX? This is helpful.

0

2件の返信を表示中 - 1 - 2件目 (全2件中)