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

Longevity of a strategy

52 replies

マイキー

顧客、bbp_participant、コミュニティ、856件の返信。.

プロフィールを表示

10年前 #114641

Since there’s no view on how long a strategy might remain profitable, and the goal of everyone here is to make money, would it not be better to push a profitable strategy to high risk, rather than run it at low risk, only to find it loses it edge over months and years?

So if you have faith in the strategy and it look good on a live account after a couple of months, ramp the risk up very high. Withdraw the original capital as soon as you have doubled your money, and then there’s nothing to lose…

Contrast running at low risk (say 2%), it would take years to make any meaningful money without a very large initial deposit, with risk of failure due to the long timescale.

0

マイキー

顧客、bbp_participant、コミュニティ、856件の返信。.

プロフィールを表示

10年前 #134994

I’m not that interested myself in trading for 5 years to make some money. I’d rather select a few strategies that look very good, then trade them live for a few months and if things look good take the risk up to the maximum point where historical drawdon is about 50% and compound the gains very quickly to make some very good money.

 

I know others have different views, like trading low risk for many many years, making a bit of money, but to me trading is about making very good money in relatively short periods of time.  Otherwise you are like spending your whole life looking at data, testing, trading, writing code for peanuts.

0

geektrader

顧客、bpb_participant、コミュニティ、503件の回答。.

プロフィールを表示

10年前 #134997

Hehe, also a good approach, as long as it makes you money + happy, it´s exactly the right way for you! Look, if there was a magic trick that made me money, I would be using that instead of spending hours figuring out strategies:) Have you made money yet with that idea?


🚀 自動化された外国為替戦略開発であなたの優位性を引き出せます 🚀

1987年から開始された歴史的な外国為替データ。28ペア、M1、99%エラーなし、生涯無料アップデート

0

パトリック

顧客、bbp_participant、コミュニティ、422件の返信。.

プロフィールを表示

10年前 #134998

You basically mean: generating a strategy and not touching it for a year, then run another backtest on that year and if it was profitable you trade it? Yes, that´s a real OOS then. Just that you had to wait a year for that to happen and got 1 year older.

 

Yes, less DMB for sure, especially if statistical enough trades (2000+, better 4000+) and lower timeframes (more data points, lowering chance to come from DMB even more). Still no guarantee for profit of course, but as close as you can get.

you dont have to wait, you just do NOT use the last year for generating. so you will use only data till 31.12.204 and year 2015 will be OOS2

0

geektrader

顧客、bpb_participant、コミュニティ、503件の回答。.

プロフィールを表示

10年前 #135002

It´s still already known data in that case and what do you do if the OOS for that one year looks bad? Right, you would never trade such a strategy and you´d either throw it away or keep on optimizing it until the 1 year OOS is profitable too, just like the IS range. And that´s effectively the same as if you just had optimized for ALL years right from the start, just that this would have been much easier:)


🚀 自動化された外国為替戦略開発であなたの優位性を引き出せます 🚀

1987年から開始された歴史的な外国為替データ。28ペア、M1、99%エラーなし、生涯無料アップデート

0

geektrader

顧客、bpb_participant、コミュニティ、503件の回答。.

プロフィールを表示

10年前 #135009

Yep:) What I am using instead of OOS, as mentioned, are bootstrapped random data sets (data created from the real data using a bootstrapping procedure with replacement) of the underlying symbol which I create via custom Python script. That is true “OOS” in the sense that it is like you live trade your strategy and you can´t optimize on that data as each set will be totally different in terms of price curve, just the main behavior of the original symbol is simulated (mathematicians know what I mean) for the reason that I can generate endless new data-sets, each one is completely different, yet each one contains the “characteristics” of the original data series of the symbol.


🚀 自動化された外国為替戦略開発であなたの優位性を引き出せます 🚀

1987年から開始された歴史的な外国為替データ。28ペア、M1、99%エラーなし、生涯無料アップデート

0

しきい値

カスタマー、bbp_participant、コミュニティ、722件の返信。.

プロフィールを表示

10年前 #135011

Quandl just posted this up today.
https://www.quandl.com/blog/interview-with-a-quant-part-one
 

Recently, Quandl interviewed a senior quantitative portfolio manager at a large hedge fund.
 

What steps do you include in your formal research process?

Early on, my biggest fear is data contamination. History is a limited resource; once you’â„¢ve run out of historical data to test against, you can’â„¢t generate any more. I’â„¢m paranoid about not exhausting my supply of uncontaminated out-of-sample data.

So I start by dividing my historical data into non-overlapping chunks. I then randomize so that even I don’â„¢t know which chunk is which. (This guards against subconscious biases: for instance, being risk-averse when I know my test dataset is 2008, or being risk-seeking in 2009).

I designate one chunk as my calibration set. I usually use Python for calibration: I use their built-in optimization libraries and have written a few of my own. In this particular example, my parameters are constrained and correlated. So I use a 2-step optimization process called the EM algorithm. Optimizers can be sensitive to initial conditions, so I use Monte Carlo to choose a number of starting points in the solution space. All of this is quite easy to do in Python.

The result of this calibration should be a set of ‘model parameters ‘“ numerical values ‘“ that can be combined with actual market observations to predict other market prices.

Once I’â„¢ve calibrated the model, I test it out of sample.

 

So, you strictly separate in-sample and out-of-sample; you blind yourself to date ranges; you use Monte Carlo to avoid starting-point biases; and you try various robustness tricks. What else do you do to ensure that you’â„¢re not fooling yourself?

I place a very high premium on parsimony. If my model requires too many parameters or has too many degrees of freedom, it’â„¢s just curve-fitting; not a model at all. So I’â„¢m constantly trying to remove factors. If the model keeps working (and remains ‘rich) with multiple factors removed, then it’â„¢s probably a good one.

A second proof of robustness is if the model works well no matter what trading strategy you build on top of it. If you can only make money using a complex non-linear scaling rule with all sorts of edge conditions, then that suggests a lack of robustness.

Finally, there’â„¢s no substitute for data. I think of every possible out-of-sample dataset that I can plausibly test the model on: different countries, different instruments, different time frames, different date frequencies. The model has to work on all of them; else you have selection bias in the results.

 

0

ステアノ

カスタマー、bbp_participant、コミュニティ、379件の返信。.

プロフィールを表示

10年前 #135016

You never know how the strategy will do “tomorrow” and it´s also clear that a portfolio does better than a single strategy in terms of safety, at least that´s my experience. I never have the expectation that the strategy is SO good that it will most likely do totally great the next days or weeks. You can never know this before hand, it´s by random chance and any feeling towards “this must be good comings weeks” is just a human hope that it will do well. Statistically that´s just a wrong assumption. A strategy can fail on day #1, do bad for the next weeks and good the following weeks, or do good from the beginning. But I would never risk a lot of money on a trade because I believe it will go well. If I´d do that, I could do manual trading, which I don´t want at all. Personally I stick to strict money management rules and workout how much of a drawdown the combined portfolio has created in the worst case in the whole backtest, then muiltiply that x2 and then calculate the % risk per trade I am going to trade to hit the historical max DD adapted to my own drawdown tolerance (have made an extra script for that outside of SQ). Then go ahead and trade it, strictly by the rules, with no intervention.

Will you expound a little more on your risk per trade formula? I am with you for 2x wc drawdown of the portfolio. Then what are the Calc steps to get amount per trade? If you don’t mind sharing.
ステアルノ

Tapatalkを使用してHUAWEI MT7-TL10から送信

0

パトリック

顧客、bbp_participant、コミュニティ、422件の返信。.

プロフィールを表示

10年前 #135024

Threshold: I know at Indexes the traders want the strategy to work on different currencies, but is it the same at FX market? i have strategies that work at one pair only..and they “work”

0

パトリック

顧客、bbp_participant、コミュニティ、422件の返信。.

プロフィールを表示

10年前 #135025

It´s still already known data in that case and what do you do if the OOS for that one year looks bad? Right, you would never trade such a strategy and you´d either throw it away or keep on optimizing it until the 1 year OOS is profitable too, just like the IS range. And that´s effectively the same as if you just had optimized for ALL years right from the start, just that this would have been much easier:)

 

i have 決して〜ない 最適化 何でも strategy. what comes from GA that i use. no neccessary to improve. when it not works on OOS 2 then i dont use it. 

0

しきい値

カスタマー、bbp_participant、コミュニティ、722件の返信。.

プロフィールを表示

10年前 #135041

Threshold: I know at Indexes the traders want the strategy to work on different currencies, but is it the same at FX market? i have strategies that work at one pair only..and the “work”

Me too.
From my standpoint I think certain pairs and indexes and commodities each have their own personalities so I’ve built a few systems for individual pairs. Currently I’m building a couple models for the S&P that I will use to take discretionary signals in the USDJPY with. Those have very specific correlated personalities. Many other pairs have their own personalities too like AUDNZD, EURGBP, etc. But my main system is the D1 ADX system I made in EA Wizard that will work on anything going back to the 1970s. Its stagnation period is brutal though so its up to your personality if you can handle that. Its also a single optimization with no WF because of MT4’s limited optimization testing. I plan to re-do it in SQ4’s platform in a single portfolio-optimization run (1 optimization for all pairs).

Many professional traders build systems that strictly specialize in the S&P, or EURUSD, or Crude. Kevin Davey has specialized systems just using data back to 2003. And Andrea Unger who got 1st 3 consecutive years in a row with 100+% annual returns in the Futures Trading Championship also uses specialized systems. In fact he doesn’t even use indicators. He hates them. He only uses time and OHLC/highest highs/ lowest lows/candle size etc.

Others like large portfolio manager Jerry Parker uses strictly systems that will only work on everything with max data going back 100+ years.

I personally like geektrader’s use of larger data as I’ve stated this was my view since I’ve first been a member here. More data is always better, though I may disagree on OOS and WFM usage, large data suits my philosophy that there are old patterns in the market that will never go away. There are new ones after the 2003 digital era too for finding.

The lesson is there isn’t a right way or a wrong way. Figure out what works for you.
 

0

geektrader

顧客、bpb_participant、コミュニティ、503件の回答。.

プロフィールを表示

10年前 #135046

Yep:) What I am using instead of OOS, as mentioned, are bootstrapped random data sets (data created from the real data using a bootstrapping procedure with replacement) of the underlying symbol which I create via custom Python script. That is true “OOS” in the sense that it is like you live trade your strategy and you can´t optimize on that data as each set will be totally different in terms of price curve, just the main behavior of the original symbol is simulated (mathematicians know what I mean) for the reason that I can generate endless new data-sets, each one is completely different, yet each one contains the “characteristics” of the original data series of the symbol.

 

Just in case someone is a Asirikuy member and wants to create bootstrapped data, it´s somewhat hidden there in the realms of the big Forum:

 

first get the script here:

 

https://asirikuy.com…&p=20951#p20940

 

Then use my solution to be able to run it on M1 data (32bit Python won´t work otherwise because the RAM usage becomes to big for a 32bit process when using 29 years of 1M data):

 

https://asirikuy.com…&p=20951#p20951

 

Have fun;)


🚀 自動化された外国為替戦略開発であなたの優位性を引き出せます 🚀

1987年から開始された歴史的な外国為替データ。28ペア、M1、99%エラーなし、生涯無料アップデート

0

ステアノ

カスタマー、bbp_participant、コミュニティ、379件の返信。.

プロフィールを表示

10年前 #135066

Yep. I guess it is down to personal preference. My personality can’t cope with a long stagnation period of a single strategy or portfolio of strategies. I also need my strategies to work on at least 20 pairs, Gold, S&P, crude etc. Granted the SL/PT range may have to be adjusted for different markets but the range of all other optimizable parameters are kept the same. I am certain I lose many excellent systems by adopting this approach but my personality is such that if I didn’t adopt this approach it would be virtually impossible for me to leave the portfolio to run without second guessing it. So I suppose it is very important to ensure the model is tradable given a particular personality type.

What is apparent is the existence of a large constellation of possibilities in relation to profitable trading model development. Some of these possibilites are known and others must be still waiting to be discovered.

In short, whatever works for you.

Sent from my GT-P5210 using Tapatalk

Notch,
You say it has to work on 20 pairs. How do you test that?

You build on one pair and then you just retest on the others? How do you define that it “trades” on the other pairs? Does it have to have positive results, have a similar equity curve to the original pair?

Or do you optimize on the new pair and then test it (because settings will most likely need to be different for different pairs)?

Tapatalkを使用してHUAWEI MT7-TL10から送信

0

しきい値

カスタマー、bbp_participant、コミュニティ、722件の返信。.

プロフィールを表示

10年前 #135072

Notch,
You say it has to work on 20 pairs. How do you test that?

You build on one pair and then you just retest on the others? How do you define that it “trades” on the other pairs? Does it have to have positive results, have a similar equity curve to the original pair?

Or do you optimize on the new pair and then test it (because settings will most likely need to be different for different pairs)?

Tapatalkを使用してHUAWEI MT7-TL10から送信

SQ has “Additional data” Fields in retest and generation. During the generation process he is generating on multiple pairs and while using the same filters like: max DD (main IS)> 30 on the additional data: max DD (A.D.)> 30 so they all meet the same standards. Optimization in SQ3.8 can only do 1 pair at a time though so to make strategies for multiple pairs obviously is done in the generation process.

ファイル GUgen.pngGUgen.png

You can build and test on multiple pairs in a single build, thats what the additional data fields are for. There are also separate filters for A.D (addition data) just like I.S./O.O.S./R.T. filters.

 

0

しきい値

カスタマー、bbp_participant、コミュニティ、722件の返信。.

プロフィールを表示

10年前 #135075

Its being added to SQ4. That was one of my major requests was portfolio-level optimization.
You’ll also be able to backtest multiple strategies on multiple pairs in a single retest like a live portfolio which is extremely important for % risk strategies since each trade’s profit/loss effects the next trade’s lot size.

0

ステアノ

カスタマー、bbp_participant、コミュニティ、379件の返信。.

プロフィールを表示

10年前 #135080

Okay. Yes, understand about AD.
I guess what I am wondering is if that is fair. Because for example gbpjpy and gbpchf are very different than USDcad. So I am thinking maybe it is unrealistic to expect the same strategy, with the same profit targets and stop loss, to work to the same standards on all these different markets. I can understand the same strategy working on different markets with different parameters. What have you found on this?

(not questioning anyone or their methods, just trying to understand)

Tapatalkを使用してHUAWEI MT7-TL10から送信

0

15件の返信を表示中 - 31 - 45件目 (全52件中)

← 1 2 3 4 →