Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Statistically Infallible 11 replies

Statistically testing every indicator with my Genetic Algoritm 132 replies

A statistically rare pattern/anomaly - reversed? 9 replies

Are my 6-mth-old results statistically valid? 17 replies

What Constitutes a Statistically Significant Sample for Back Testing? 4 replies

Attachments: How to keep in faith with a statistically profitable system?
Exit Attachments

How to keep in faith with a statistically profitable system?

  • Post #1
  • Quote
  • First Post: Dec 30, 2010 11:59pm Dec 30, 2010 11:59pm
Dear fellow traders, i have a question stuck long enough and still can't figure it out by myself, maybe i didn't learn enough about probability math.
This question is appointed for any quants, please help me with this

Here we go :
Let's say we have test our system for once, say it 100 trades, and we got 90 winning trades and 10 losing trades (so we have simply 90% winning and 10% losing in percentage terms)
Then, we run the system in the live account, and suddenly,we have like 30 consecutive losses. According to our system it was 10% of failures, so we should theoretically stick with our system and expecting that we will at least have other 270 winners (if we are planning to make a 300 trade in a year), am i right? please correct if anything wrong

Nah, the question is, how could we know the next trades will be 270 consecutive winning trades? it can also be 100 consecutive losing trades then 900 winning trades in a row, and it will still be a 90% winning and 10% losing.
If so, how could we simply apply what we have tested systems, even it is a 99% winning system, it also can be a 10 consecutive losing trades and 990 winning trades, how can we keep in faith with our system that generates large numbers of losing trades in a row while it was statistically a profitable systems?

Please help me with this problem
We don't. Even after losing 30 trades in a row, we're still only 90% likely to win our next trade. The numbers will never "balance" (i.e. experience 30 losses, but don't expect 270 winners to make up for it), but the larger our sample grows, the more likely it will start to reflect our edge. Over time, those 30 consecutive losses will become a memory.

That said, losing 30 in a row is beyond horribly unlucky, and it's far easier for a 50% system to do it than a 90% system.
100 trades is way too small sample size, you cant tell anything from that.
From a strict probability and quant viewpoint you don't know if the next trades will be 270 consecutive winners.

How can you keep faith? By knowing that your backtests are truly accurate, by using the system in a live account and seeing the performance. By asking yourself why there is such a huge discrepancy between the live account and backtest performance.

Also, if someone thinks they have a system with 90% winrate and it generates 30 consecutive losses, they could attribute it to a run of bad luck. The vastly more likely scenario is that their system is not 90% winrate no matter how hard they "think" it is or how the backtest turns out. Reality and live trading with emotions and money on the line is where it really counts.

I used to run a system that had an 75-90% winrate. If it generated 1 loss I would very reluctantly kinda attribute it to bad luck. If it generated 2 losses in a row the warning lights would fire in my trading room and in my personality and I had to figure out why I was taking losses and whether I was following the rules of my system. I would never tolerate more than 2 losses in a row.

Generally if you are trading a system with such high win rates, you should know pretty exactly what causes your trades to succeed or fail. You should know why your trades succeed or fail to a fairly precise degree.
Private message me for a link to my order flow website.
Truth of the matter is, no kelly criteria can be accurate because you are always taking a tiny sample of infinity. 100 trades.... so what? What happens if you measured the 30 consecutive losers first? Would you say the expectancy is 0?

Even with a large data set (thousands of trades) no chart or entry/exit system will conform to it's historical performance..... every year's chart is different and unpredictable, and so must be the expectancy.
It would check my system again. One thing is for sure, you don't have a 90% chance for a win with this system. Depending on your style, having a high % win isn't essential to make a killing. I'm more than happy with 60% winning rate thanks to high R:R.

Below is a table showing probablity for consecutive loses within 100 trades with 90% chance for a win.
Attached Image (click to enlarge)
Click to Enlarge Name: 1.png Size: 31 KB
Measure twice, cut once
Quoting Mr J
That said, losing 30 in a row is beyond horribly unlucky, and it's far easier for a 50% system to do it than a 90% system..

Quoting grkfx
From a strict probability and quant viewpoint you don't know if the next trades will be 270 consecutive winners..

Quoting aediaz1
It would check my system again. One thing is for sure, you don't have a 90% chance for a win with this system. Depending on your style, having a high % win isn't essential to make a killing. I'm more than happy with 60% winning rate thanks to high R:R.

So what do you traders think is the maximum numbers of consecutive losses can be tolerated for a profitable trading system?
What will you do if this number reached and you still don't have any idea what is wrong with your ts?



Quoting tdion
Truth of the matter is, no kelly criteria can be accurate because you are always taking a tiny sample of infinity. 100 trades.... so what? What happens if you measured the 30 consecutive losers first? Would you say the expectancy is 0?

Even with a large data set (thousands of trades) no chart or entry/exit system will conform to it's historical performance..... every year's chart is different and unpredictable, and so must be the expectancy.
Quoting bjorken
100 trades is way too small sample size, you cant tell anything from that.
So, how can we measures one trading system's winning / losing performance? how much samples should we take?
Quoting in_drag88
So what do you traders think is the maximum numbers of consecutive losses can be tolerated for a profitable trading system?
What is more important is the seize of your loses and how you handle them because every system will naturally have losers.

Personal, my stress level for consecutive losers is about 4 or 5. More than that and I'll cut my positions (risk %) by 50%.

Important statistics for me is

  1. Average win > average loss (expectancy). Usually, my avg wins are 1.5-3 times larger than avg loses.
  2. Drawdown %. Myfxbook.com have a new statistics called Risk Of Ruin (shows the probability of losing a certain amount of your account, ranging from 100%-10%) which I think is a great tool, having enough samples to measure of course.

Measure twice, cut once
Quoting in_drag88
Dear fellow traders, i have a question stuck long enough and still can't figure it out by myself, maybe i didn't learn enough about probability math.
This question is appointed for any quants, please help me with this

Here we go :
Let's say we have test our system for once, say it 100 trades, and we got 90 winning trades and 10 losing trades (so we have simply 90% winning and 10% losing in percentage terms)
Then, we run the system in the live account, and suddenly,we have like 30 consecutive losses. According to our system it was 10% of failures,...
What is the chances of 30 consecutive losses within 300 trades for 90% winrate system, or the chances of 10 consecutive losses within 1000 trades for 99% winrate system? Not even once since big bang.

In fact the chances of at least 1 time of 10 consecutive losses within 1,000,000,000 trades for a 90% winrate system is about 10% only.

So if it does happen, It is very likely that you were making mistakes in your testing.
Quoting in_drag88
So what do you traders think is the maximum numbers of consecutive losses can be tolerated for a profitable trading system?
What will you do if this number reached and you still don't have any idea what is wrong with your ts?
I've had a period of 34 consecutive losses....and there is nothing wrong with my system. My system is, and has been since I started trading it, very profitable over the long term. The maximum number of losses that can be tolerated in a profitable system is entirely down to your strategy, money management and personal approach to risk/loss.

If you're in a losing streak and as you put it you "still don't have any idea what is wrong with your ts", then I would imagine that the real problem is that you don't really know how your system works or what is normal or expected in terms performance. As others have stated above, 100 trade sample size is most likely way too small. Why does there have to be anything wrong with your system, just because you're in a losing streak?

Trading is a paradox between the small numbers and large. If you had a 90% win rate, then you could expect over a series of large numbers (10,000, 100,000, ...more) that the statistics would balance out to give you approx 90 wins for every 10 losses. BUT, on a small scale, i.e. each individual trade, anything could happen. Individual trades are pretty much random and 50/50. AND...each individual trade is completely independent of the last or the next. Just because you've had 10 losses in a row, there's nothing to say the next trade or series of trades is likely to be a win or winning streak. You just have to trust over the long term the law of large numbers balances out.

Just like a coin flip with a 50% heads to tails ratio. If you flip 10 heads in a row, what are the odds that the next flip with be heads again? Surely the odds say the next flip is more likely to be tails! Well the odds on the next flip is still 50%. Every coin flip is independent of the last, and every flip is always 50/50 regardless of the preceding series of flips.

If you've genuinely tested your strategy thoroughly and accurately and you are 100% confident that your win/loss rate is correct, and you have a positive expectancy over the long term, then a losing streak shouldn't make any difference at all. If you've done your research properly, a losing streak should come as no surprise and should be absolutely no reason to question or doubt your system. If the losses are making you doubt your system, then I think (and this is only my opinion) you haven't done enough ground work before starting to trade.

I trust my strategy 100%. When I get a big losing streak it is simply reinforcing what I already believed to be normal and expected and part of my winning system. I expect to hit big losing streaks. And that gives me the confidence to keep following my rules, which over the long term balances everything out (law of large numbers) and I finish each year with a healthy profit. If I thought a losing streak was anything other than normal, I would be constantly jumping from one system to another and losing year after year like so many do.

You may need to do some more study and research to gain the confidence to know what to expect over a large sample of trades using your system. Your system may be fine, and the losing streak you encounter is 100% normal and nothing for you to be concerned about.
Quoting in_drag88
Here we go :
Let's say we have test our system for once, say it 100 trades, and we got 90 winning trades and 10 losing trades (so we have simply 90% winning and 10% losing in percentage terms)
Then, we run the system in the live account, and suddenly,we have like 30 consecutive losses. According to our system it was 10% of failures, so we should theoretically stick with our system and expecting that we will at least have other 270 winners (if we are planning to make a 300 trade in a year), am i right? please correct if anything wrong
You only approach the 90% and the 10% when the sample size is infinite. Let me put it like this, as much as you increase the sample, the win and loss probabilities will approach the 90% and the 10%. With smaller samples, there will always be a tendency for marginal error, so the win and loss percentages will not be 90% and 10%, they will be something else.

Get a coin and toss it 10 times and see how many heads and how many tails do you get. Then toss it 100 times and see how many heads and how many tails do you get, etc....You will notice that as you increase the sample, the probabilities approach the theoretical value of 50% and 50%.

Quoting in_drag88
Nah, the question is, how could we know the next trades will be 270 consecutive winning trades? it can also be 100 consecutive losing trades then 900 winning trades in a row, and it will still be a 90% winning and 10% losing.
If so, how could we simply apply what we have tested systems, even it is a 99% winning system, it also can be a 10 consecutive losing trades and 990 winning trades, how can we keep in faith with our system that generates large numbers of losing trades in a row while it was statistically a profitable systems?

Please help me...
Now, that's an interesting question. You won't know for sure, that's why you will have to stress test your system against potential drawdowns and adjust your position sizing accordingly to prevent the risk of ruin.

Your backtest results must include the max. drawdown experienced in the past as well as the max number of consecutive losing trades in a row.

It's up to you to take such drawdown and max number of consecutive losers in a row as a reference for the future or even to apply a more severe factor to prepare for a worse situation in the future.

If you need to read more on this topic, I would highly recommend this book "Way of the Turtle" by Curtis Faith.

http://www.amazon.com/Way-Turtle-Met...3803385&sr=8-1

That's a great book and I would really recommend it to anyone who is concerned about what should be done after you have already tested your system. There is usually a lot of work to be done after that.

Best Regards,

Nader
This is a prime example of the failure of math in the market. I absolutely agree that 100 trades is a very small sample. However, even testing the system over the past several years will not produce a reliable result. Why?

Because the market is made up of people not math. Each and every moment in the market has an infinite number of possibilities influencing the outcome of any given tick, much less the direction of the market.

I am someone who believes in learning to "read" trader sentiment through price action and have learned that even that will change like a fickle high school girl in the back seat of a chevy. The only thing that can increase you "probability" of success over time and/or over the course of any given trade is risk management.

My advise:

  1. Take a larger sample of trades (min 4 years)
  2. Find the worst draw down in a single trade
  3. Adjust your position size so that worst drawdown is 1% of your account
  4. Set a stop that is twice that drawn down so your max risk is 2%
  5. Take profit on each and every winner (scale out or close out) preferably more than your risk
  6. FORGET about trying to "forsee" market direction
  7. If you losses exceed 6% in any given month STOP TRADING for that month! Go back and re-evaluate the efficacy of your system or method.


Don't over complicate you system or try to make it "match" past data. Its tomorrow's data that will kick your butt and you don't know what that will be until its too late.

Bottom line? As the saying goes, "...cut your losers short, and let your winners run"

Also, it seems like your SL are too small or have no technical significance (if you use TA). Hiding SL outside the daily range will give you statistical advance.
Measure twice, cut once
Quoting gspajon
This is a prime example of the failure of math in the market. I absolutely agree that 100 trades is a very small sample. However, even testing the system over the past several years will not produce a reliable result. Why?

Because the market is made up of people not math. Each and every moment in the market has an infinite number of possibilities influencing the outcome of any given tick, much less the direction of the market.
gspajon, I will have to slightly disagree. I do agree that the market is made up to people and not math and that those people affect each and every tick and that hundreds of thousands of possibilities can be expected every tick.

However, when you try to model something based on the collective effects of several inputs, you will have to use math and probabilities. Since we agree that the price (Over several time frames and from the tick to years) is a collective output of hundreds of thousands of peoples' actions and since we can't study the collective actions of market participants, we try to model trading systems that incorporate peoples' actions that happened in the past based that it will be repeated again in the future!!! Does it make any sense?

It gets a lot complicated to really assess if the system you are looking at is really reflecting and doing what it is supposed to do or is it just another rubbish nonsense system or model?

It also gets more complicated when you try to figure out the different possible outcomes given the change in variables of your system and how you really should respond to them. This is when you start putting your system test results under inspection and serious testing to see if it's really robust or not, because the future is not going to be 100% exactly like the past. A 1% deviation in future conditions from past conditions can completely destroy your system if it's not a robust one.

A robust system is defined as a system that can withstand future conditions that are not exactly like past conditions yielding results that do not deviate too much from historical testing results.

When you try to put your system under testing to know if its' robust or not, you usually use things like Monte Carlo simulations in order to get a feeling of your system robustness and to make sure that the system is not going to breakdown if the future unfolds to be very slightly different than the past.



Quoting gspajon
I am someone who believes in learning to "read" trader sentiment through price action and have learned that even that will change like a fickle high school girl in the back seat of a chevy. The only thing that can increase you "probability" of success over time and/or over the course of any given trade is risk management.
Risk management can be incorporated inside system trading as well.

Quoting gspajon
My advise:

  1. Take a larger sample of trades (min 4 years)
  2. Find the worst draw down in a single trade
  3. Adjust your position size so that worst drawdown is 1% of your account
  4. Set a stop that is twice that drawn down so your max risk is 2%
  5. Take profit on each and every winner (scale out or close out) preferably more than your risk
  6. FORGET about trying to "forsee" market direction
  7. If you losses exceed 6% in any given month STOP TRADING for that month! Go back and re-evaluate the efficacy of your system or method.

4 years is not enough. A minimum of 10 Years is advisable.

If you adjust your position size so that the drawdown is 1% of your account, you are also hindering and limiting your profit potential.


When I said I have to disagree with you, I said so because a lot of success stories in real world proved that system trading does work, however, it's not only about the system, there is much more than that. Take the turtle traders for example. These guys blindly followed the rules, despite that the rules were flexible enough to give them some room to maneuver, yet the majority of them were very successful.

Best Regards,

Nader
Quoting in_drag88
Dear fellow traders, i have a question stuck long enough and still can't figure it out by myself, maybe i didn't learn enough about probability math.
This question is appointed for any quants, please help me with this

Here we go :
Let's say we have test our system for once, say it 100 trades, and we got 90 winning trades and 10 losing trades (so we have simply 90% winning and 10% losing in percentage terms)
Then, we run the system in the live account, and suddenly,we have like 30 consecutive losses. According to our system it was 10% of failures,...
Don't test in Demo. No matter what anyone says, it's NOT the same. You will always get different results between demo and a live account. And it has NOTHING to do with psychology or going live jitters.

If you are testing a win/loss ratio, do it live on micro account with very small lots.
Quoting KumoDragon
Don't test in Demo. No matter what anyone says, it's NOT the same. You will always get different results between demo and a live account. And it has NOTHING to do with psychology or going live jitters.

If you are testing a win/loss ratio, do it live on micro account with very small lots.
You can replicate much of live's psychology in demo, with practice. Why invoke taxes? There's a world of room in paper, simulator and demo to hone one's technical edge before testing with pennies.
Quoting in_drag88
Here we go :
Let's say we have test our system for once, say it 100 trades, and we got 90 winning trades and 10 losing trades (so we have simply 90% winning and 10% losing in percentage terms)
Read a statistics book, 100 trades is definitely not enough. Try 10,000 trades, the minimum sample size for any statistical relevance. I would not have faith in a system I only tested over 100 trades.
Quoting in_drag88
Dear fellow traders, i have a question stuck long enough and still can't figure it out by myself, maybe i didn't learn enough about probability math.
This question is appointed for any quants, please help me with this

Here we go :
Let's say we have test our system for once, say it 100 trades, and we got 90 winning trades and 10 losing trades (so we have simply 90% winning and 10% losing in percentage terms)
Then, we run the system in the live account, and suddenly,we have like 30 consecutive losses. According to our system it was 10% of failures,...
Hello

today i was listining to an video from a guy how work by fxdd and he sayd that people when they trade for a long time a demo account then then will not feel the risk so they will trade with out emotion if they do that for a long time so maybe that is the case.

because when you start live trading you start use your emotion with where you place your stoplosse or take profit and when to exit the trade.

I hope this helps maybe that is some think else at all but i wrote that so that you can be aware of that ulso.

greet you and talk to you in 2011,

Yoenes
just work out what the expactancy is. Also, focus on the process rather than outcome. the numbers will work for you in the long run.

this is how poker players keep their money and make more and the same thing great traders will tell you. just watch broke: the new american dream. the traders in there who have made millions will tell you that and so will great poker players who have made millions. makes all the sense to me. no need to overcomplicate this and try to say, "naw that doesnt work blah blah blah" maybe when you get to be a millionare off of this, then you can tell those guys who actually make the money different.
Pound it through your skull... a "winning percentage" is a lie. The minute you set an expectancy in your mind, you are imposing yourself on the market to do what you expect it to do.

This is a fallacy most folks make--- the market never conforms to anyone.... it is independent, and thus, will always act on it's own intrinsic behalf.
0 traders viewing now