A quantitative approach to trading Forex and Forex Derivatives | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Quantitative and Algorithmic Trading 188 replies

Quantitative Approach to Backtesting 24 replies

My approach to forex trading 44 replies

Gill Blake's approach to forex trading 5 replies

The Shark's Trading Book - Quantitative Trading 2 replies

Attachments: A quantitative approach to trading Forex and Forex Derivatives
Exit Attachments

A quantitative approach to trading Forex and Forex Derivatives

  • #21
  • Sep 12, 2017 7:20am Sep 12, 2017 7:20am
Quoting stt
In the quant analysis, one of the most important question is how do you avoid data overfitting bias - there are hundreds of articles on this (pretty much every stats discussions leads here) - check the discussion here https://www.reddit.com/r/algotrading/comments/6febbf/dangers_of_backtest_overfitting_marcos_lopez_de/ So my questions is how to you make sure the correlations you are finding are not spurious by chance and have some real predictive power. Thanks!!
Very good question. I have what i call a design set of data, which might be say 1 year on which i tweak my parameters, and then i test it against a lot more data without tweaking
I live entirely from the growth of my personal trading account
  • #22
  • Sep 12, 2017 7:28am Sep 12, 2017 7:28am
Quoting GuyBaker
{quote} Very good question. I have what i call a design set of data, which might be say 1 year on which i tweak my parameters, and then i test it against a lot more data without tweaking
Great. That's the cross validation method where you have separate training and test data. As you realize if you do this 10 (or even 100s) of time, the integrity of this method breaks as now the tests data is essentially part of training data.

This is the thing Marco De Prado discusses in his paper - the most iterations you test, the lower your confidence on the model as over fitting bias creeps in.
Think : Trade : Live Life
  • #23
  • Edited 9:00am Sep 12, 2017 8:37am | Edited 9:00am
Quoting stt
{quote} Great. That's the cross validation method where you have separate training and test data. As you realize if you do this 10 (or even 100s) of time, the integrity of this method breaks as now the tests data is essentially part of training data. This is the thing Marco De Prado discusses in his paper - the most iterations you test, the lower your confidence on the model as over fitting bias creeps in.
If you tweak enough parameters even random data can appear to show a pattern, I learned about this the same way I seem to learn everything in life is by making the mistake myself
I live entirely from the growth of my personal trading account
  • #24
  • Sep 13, 2017 1:27pm Sep 13, 2017 1:27pm
What sort of margin do you use? Account size? I assume you are a scalper
  • #25
  • Sep 13, 2017 2:17pm Sep 13, 2017 2:17pm
Quoting Griff90
What sort of margin do you use? Account size? I assume you are a scalper

No I am not a scalper all of my trades are executed by my proprietary software which runs on servers in London, for the system I have talked about on here I use 1% of my available cash per trade, so at this time I am risking about £2150 per trade.
I live entirely from the growth of my personal trading account
  • #26
  • Sep 13, 2017 2:27pm Sep 13, 2017 2:27pm
Quoting GuyBaker
{quote} No I am not a scalper all of my trades are executed by my proprietary software which runs on servers in London, for the system I have talked about on here I use 1% of my available cash per trade, so at this time I am risking about £2150 per trade.

Thank you.

Okay, I know you have said a few times you will not go into great details about some things, could you beat around the bush and talk about how you do trade? Whether it be via a certain set of indicators or naked trading, if its just one system you use or have a few automated systems you use. Is it currency you trade or other instruments?
  • #27
  • Sep 13, 2017 2:40pm Sep 13, 2017 2:40pm
Quoting Griff90
{quote} Thank you. Okay, I know you have said a few times you will not go into great details about some things, could you beat around the bush and talk about how you do trade? Whether it be via a certain set of indicators or naked trading, if its just one system you use or have a few automated systems you use. Is it currency you trade or other instruments?
I used a quantitative approach to build my black box, the inner workings of that box only I know. I have used price data of course, some indicators and news releases as per the calendar on this site. It is the fundamentals that move the market most of the time. The system detailed on here trades only Forex fixed price options, I have another in trail which trades the EUR/USD
I live entirely from the growth of my personal trading account
  • #28
  • Sep 14, 2017 2:07am Sep 14, 2017 2:07am
For those that may be interested here is a graph showing my return in 1% units since 1/12/15, I have improved the system as I have gone along and you will notice that the growth line has become smoother over time. I find it very useful to record every trade executed in an Excel sheet at the end of each day as this accumulates a resource that can be analysed to test new ideas which might improve the performance of the system.
Attached File(s)
File Type: pdf chart.pdf   117 KB | 463 downloads
I live entirely from the growth of my personal trading account
  • #29
  • Sep 17, 2017 12:49pm Sep 17, 2017 12:49pm
Quoting GuyBaker
{quote} No I am not a scalper all of my trades are executed by my proprietary software which runs on servers in London, for the system I have talked about on here I use 1% of my available cash per trade, so at this time I am risking about £2150 per trade.
only 1 open trade at a time? if not, second trade would be 1% or (99% of available cash) and so on?
Staying in my lane...
  • #30
  • Sep 17, 2017 1:13pm Sep 17, 2017 1:13pm
Quoting stt
{quote} Great. That's the cross validation method where you have separate training and test data. As you realize if you do this 10 (or even 100s) of time, the integrity of this method breaks as now the tests data is essentially part of training data. This is the thing Marco De Prado discusses in his paper - the most iterations you test, the lower your confidence on the model as over fitting bias creeps in.
Both of you have some experience with overfitting and training so I will share what I am 'trying' to do to introduce machine learning in my automation.

  1. By definition, I have broken down all market conditions into 4 up phases and 4 down phases in a trend. (I consider range also as a (micro)trend
  2. With this universal approach, I 'data-mine' 20-80 million price points depending on the testing range of historical data to generate 5K-8K raw trade signals which go thru another confluence ringing process to get me under 500 or so "filtered" trade signals across a select combination of 28 pairs and select timeframes
  3. Next step is to analyze which of these signals gained profit. this process bring it down to under 200 or so signals. This is now my "Training" footprint. Every "Training Period", this list is evaluated with forward testing data and merged/consolidated/filtered for use in the following "Period". Currently, "Training Period" is at the end of each month (for the last 3 months of test/train data) as it takes about 3 days to run my mql script. It has about 8 nested for loops lol)
  4. I don't mine / store the data in a database which could speed things up. This is all CPU processing power on my laptop running at 90+% for 3 days to crunch the numbers and filter out the bad apples from the good ones.

Has someone tried this approach successfully? Some weeks, my trades in forward testing work like MAGIC increasing the win rate further from my modelling. i.e. 70% winrate in my model turns to 85-90% win rate. but some weeks, it fails miserably and cuts the winrate to more than half. This happens only when there is no/low volatility for 2-3 consecutive days which I knew can and will happen. I don't use ANY indicator or calculation or formula in my automation.

Any thoughts on this approach? I can't be the only one doing this :-)

EDIT1: I forgot to mention.... signals can happen at any time as as long as I have available margin I take those signals. With that said, I do have many controls in my EA to prevent over trading, over exposure etc. examples include minimum space between trades per pair/tf, add only if previous trades are in profit, exit at close etc. Doing all this does make me 'skip' signals so perhaps I am skipping good signals and taking bad signals. I have no control over it when I try on a smaller account in demo. On a LARGE $1+M demo account, forward testing works beautifully against my model as I am able to take ALL my signals. This is my conundrum at the moment. my model requires up to 500 open trades and I ignore wicks so need a large pad to accommodate floating equity fluctuations across 500+ trades.

Staying in my lane...
  • #31
  • Edited 3:07pm Sep 17, 2017 2:38pm | Edited 3:07pm
Quoting VEEFX
{quote} only 1 open trade at a time? if not, second trade would be 1% or (99% of available cash) and so on?
There can be multiple trades open simultaneously , 1% is a simplification, in reality I set my unit size to 1% of my available cash at the beginning of each month and then I increase it at the beginning of the next month if the balance has risen otherwise I leave it the same, so actually because most months are profitable most of the time each trade is <1% of my cash.
I live entirely from the growth of my personal trading account
  • #32
  • Sep 17, 2017 2:56pm Sep 17, 2017 2:56pm
Quoting VEEFX
{quote} Both of you have some experience with overfitting and training so I will share what I am 'trying' to do to introduce machine learning in my automation. By definition, I have broken down all market conditions into 4 up phases and 4 down phases in a trend. (I consider range also as a (micro)trend With this universal approach, I 'data-mine' 20-80 million price points depending on the testing range of historical data to generate 5K-8K raw trade signals which go thru another confluence ringing process to get me under 500 or so "filtered" trade...

I use a similar approach, however, I code one back test platform which includes all of the filters in one pass, the logic in my test platform effectively becomes my execution logic. What language are you using? MT4? I don't use MT4, I prefer to have control of everything even the calculation of any indicators in my own code, I am using Excel with VBA, I build as many of the calculations into the sheet as possible and then use a compiled VBA script to do anything that is not practical to do on the sheet. Even my execution platform runs on Excel. In my test platform I also keep a track of maximum draw down from peak which I use to calculate how much I can reasonably risk per trade. My win rate over the long term is 60% but it can vary wildly from day to day generally over a 100 trades or so it is in the region of 60%


This bit of your post "This is my conundrum at the moment. my model requires up to 500 open trades and I ignore wicks so need a large pad to accommodate floating equity fluctuations across 500+ trades." if I am honest having that many trades open at any one point seems impractical, can your broker even handle that? The system I have running which is described in this thread rarely has more than 3 open at one time and the other system which is in test is capped at 10.
I live entirely from the growth of my personal trading account
  • #33
  • Edited 3:55pm Sep 17, 2017 3:22pm | Edited 3:55pm
Quoting GuyBaker
{quote} There can be multiple trades open simultaneously , 1% is a simplification, in reality I set my unit size to 1% of my available cash at the beginning of each month and then I increase it at the beginning of the next month if the balance has risen otherwise I leave it the same, so actually because most months are profitable most of the time each trade is <1% of my cash.
Great. I use x% of MathMax (Equity @ start of week, Equity @Start on month, AccountBalance(), Max Equity Peak from start of week/month) in my %Risk and %Profit taking. It helps with exponential growth (and recovery from drawdown)
Staying in my lane...
  • #34
  • Sep 17, 2017 3:37pm Sep 17, 2017 3:37pm
Quoting GuyBaker
{quote} I use a similar approach, however, I code one back test platform which includes all of the filters in one pass, the logic in my test platform effectively becomes my execution logic. What language are you using? MT4? I don't use MT4, I prefer to have control of everything even the calculation of any indicators in my own code, I am using Excel with VBA, I build as many of the calculations into the sheet as possible and then use a compiled VBA script to do anything that is not practical to do on the sheet. Even my execution platform runs on...
MT4 will do for now but eventually it will have to be via API on a server that I will host and maintain :-)
500 is not that realistic as my primary objective is to have "shock absorbers" to accommodate any news or market condition. I cannot rely on STOPS being triggered as they are not gauranteed. There are many inherent risks that are out of our control so having a portfolio with balanced exposure should, in theory, handle any volatility. Mine ignores wicks across all pairs and timeframes for this very reason. How much I make is not important at all to me. Even 0.1% per week is a huge accomplishment as long as other criteria are met.
Staying in my lane...
  • #35
  • Edited 2:54am Sep 18, 2017 1:58am | Edited 2:54am
Quoting VEEFX
{quote} MT4 will do for now but eventually it will have to be via API on a server that I will host and maintain :-) 500 is not that realistic as my primary objective is to have "shock absorbers" to accommodate any news or market condition. I cannot rely on STOPS being triggered as they are not gauranteed. There are many inherent risks that are out of our control so having a portfolio with balanced exposure should, in theory, handle any volatility. Mine ignores wicks across all pairs and timeframes for this very reason. How much I make is not important...

Yeah I get the principal, 500 just feels unwieldy to me, I guess 500 trades in the whole market or even for a broker is a drop in the ocean, I do wonder if a given retail broker's trade management software is geared up for that many trades open for just one user though, have you done any tests, obviously in demo mode?
I live entirely from the growth of my personal trading account
  • #36
  • Sep 18, 2017 1:52pm Sep 18, 2017 1:52pm
Quoting GuyBaker
{quote} Yeah I get the principal, 500 just feels unwieldy to me, I guess 500 trades in the whole market or even for a broker is a drop in the ocean, I do wonder if a given retail broker's trade management software is geared up for that many trades open for just one user though, have you done any tests, obviously in demo mode?
Some limit and some allow unlimited as long as the EA does not flood their servers. Mine only will open few trades a day. This is just based on my portfolio modelling. I have tried it on demo and it work extremely well during volatile times absorbing most of the market shocks.... actually market shocks are preferred as it kicks off my profit manager module instantly. but demo means squat so trying to muster the strength to try this model out in live with $20K starting capital. hopefully, I will get there and consider this truly a hands-free operation :-)
Staying in my lane...
  • #37
  • Sep 19, 2017 2:51am Sep 19, 2017 2:51am
Quoting VEEFX
{quote} Some limit and some allow unlimited as long as the EA does not flood their servers. Mine only will open few trades a day. This is just based on my portfolio modelling. I have tried it on demo and it work extremely well during volatile times absorbing most of the market shocks.... actually market shocks are preferred as it kicks off my profit manager module instantly. but demo means squat so trying to muster the strength to try this model out in live with $20K starting capital. hopefully, I will get there and consider this truly a hands-free...

Very interesting approach, what do you expect as an annual return? What likely max drawdown?
I live entirely from the growth of my personal trading account
  • #38
  • Sep 19, 2017 4:00pm Sep 19, 2017 4:00pm
Quoting GuyBaker
{quote} Very interesting approach, what do you expect as an annual return? What likely max drawdown?
In my modelling, I am not at all focused on returns (weekly/monthly or annual... I know it sounds retarded but I can generate non-consistent returns by manual trading from events so not a priority for my automation). What's my priority?
1) to limit the drawdown to under 15% max. 2) Consistency in execution via automation 3) Enforce discipline via automation 4) Conduct business that I cannot do it manually 5) Freedom from the screen and chart analysis.... chartless trading!.

I cannot and will not put any capital at risk unless I am 1000% confident it can absorb all market shocks as I have come to realization that you need capital with moderate leverage to make decent money in FX. High risk, High Reward is good for hobby but not sustainable in the long run. 1:50 leverage is plenty for me. Oanda is okay for me. Non-Hedging is also okay for me (my profits increase significantly on Hedged accounts versus non-hedged!). FIFO is a challenge but Oanda allows some accommodation with varying lot size. safety of funds is also okay with Oanda.

Now, here's something that kicks me in the gut. On top of the 500+ open trades with my model, I also need 3 to 5 times the number of pairs * number of timeframes of available margin to run my own "ponzi scheme". Why? because, when I realize profits, it messes up the "model" portfolio momentarily. Sure... Profit taking frees up margin so I have some room to wiggle but during this wiggling and waiting to accept new signals (I dont trade below H1), I need that cushion as I am following macro and micro trends all the time in a universal and market neutral way.... regardless of direction with the outcome towards optimization driven by market forces/cycles. If a flash crash were to occur during this "wiggling", I am at risk of imbalance and getting overexposed. That is why I stop opening positions at a certain margin level and I start closing max profit, max loss, max profit, max loss in a sequence when margin drops below a certain level. Doing this absorbs market volatility ... IN THEORY.

EDIT - made some changes
Staying in my lane...
  • #39
  • Edited Sep 20, 2017 1:14am Sep 19, 2017 11:59pm | Edited Sep 20, 2017 1:14am
Quoting VEEFX
What's my priority? 1) to limit the drawdown to under 15% max. 2) Consistency in execution via automation 3) Enforce discipline via automation 4) Conduct business that I cannot do it manually 5) Freedom from the screen and chart analysis.... chartless trading!.

My criteria exactly! Although I never started making money from trading on anything like a consistent basis until I embraced automation fully.

The system I currently live from etc, has a theoretical max draw down of 15% and now a monthly expected average return of 8.6%, your concerns about risk and flash crashes were in my mind also when I designed this system, I limit my risk primarily by using fixed priced options contracts, the are not fashionable, indeed I have been accused of being a scammer for even talking about them! But they do mean my exposure to the market is extremely controlled, indeed most of the time my account sits in cash, and even when I have contracts open flash crashes etc are not a concern because of the nature of the contracts.

I do not actually use this exchange at the moment, but it is interesting https://www.nadex.com/
I live entirely from the growth of my personal trading account
  • #40
  • Sep 20, 2017 1:45am Sep 20, 2017 1:45am
Quoting GuyBaker
{quote} My criteria exactly! Although I never started making money from trading on anything like a consistent basis until I embraced automation fully. The system I currently live from etc, has a theoretical max draw down of 15% and now a monthly expected average return of 8.6%, your concerns about risk and flash crashes were in my mind also when I designed this system, I limit my risk primarily by using fixed priced options contracts, the are not fashionable, indeed I have been accused of being a scammer for even talking about them! But they do...
Thanks man. Interesting that nadex is allowed and available in US. It's regulated and segregated so definitely worth exploring further. I never dipped my fingers into options in the past. Why would anyone accuse you of being a scammer especially when you can limit drawdowns with a decent monthly return? Screw them and share your strategy for noobs like me bro :-

A brief peek and I already have an idea how to use my weekly signals with binary options as stops. My MN+W1, MN+D1 and W1+D1 have the highest winrate but comes with bigger drawdown as I ignore wicks. To circumvent, I wait for candle direction to turn in my favor before I enter. With options, I can completely eliminate drawdown on higher timeframes.
Staying in my lane...
0 traders viewing now