You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Pair Trading Strategy Journal
- Thread starter jonnysharp
- Start date
Long: AON $46.10
Short: BTU $35.75
Where is everyone? What? Did they find Jimmy Hoffa?
I'm late to the party as usual...after reading 400+ pages.
Has everyone found an easier way to make $$ rather than pair trading? Playing the divergence seems to be the path of least resistance lately, for me.
Short: BTU $35.75
Where is everyone? What? Did they find Jimmy Hoffa?
I'm late to the party as usual...after reading 400+ pages.
Has everyone found an easier way to make $$ rather than pair trading? Playing the divergence seems to be the path of least resistance lately, for me.
Quote from sals3r0:
If anyone is interested, I have created a website where you can backtest your pairs online. It uses the C++ based backtester I use for screening pairs.
Please note the website is still in BETA stage and there might be bugs. Enjoy and good luck trading. I appreciate you report bugs you find.
Address: http://www.pairtradinglab.com
Did a big update today - you can create studies. Explanation:
Studies are essentially multi-backtests performed on whole ranges of indicator periods. If you create a study for pair of stocks, it will silently create backtests for all period ranges (10-50) at the same moment. Results are then displayed using a scatter plot of Compound Annual Growth Rate (CAGR) in one axis and maximal draw-down in the another. In addition, stock prices are displayed in a chart together with multiple correlation indicators to give further insight about pair history.
Studies are powerful tools used mainly to evaluate the robustness of the pair, how those stocks hold up for various settings. Because you can actually see the behavior of the pair for all possible parameters at once, this tool can be used to help eliminating the risk of over-optimization (i.e. some pair is bad, but for some particular periods gives good results - with a single backtest, you might not notice that). In general, good robust pair should exhibit satisfying results for all indicator periods in the test instead just for few.
Example: http://www.pairtradinglab.com/studies/TwnfcIhdDeO-DT3N
Looks interesting. Haven't got it figured out yet, but obviously a great project requiring a lot of work.
Is there a way to search for highly profitable pairs, or must we bring our possibilities to the tester to confirm suspicions on profitability?
The CAPTCHAs are really hard to read. Is it necessary as frequently as it now requires -- for each click on each point on the scatter graph a new CAPTCHA must be decoded. It's a bit of a pain. Why not have a login requiring a CAPTCHA just once?
Walt B
Is there a way to search for highly profitable pairs, or must we bring our possibilities to the tester to confirm suspicions on profitability?
The CAPTCHAs are really hard to read. Is it necessary as frequently as it now requires -- for each click on each point on the scatter graph a new CAPTCHA must be decoded. It's a bit of a pain. Why not have a login requiring a CAPTCHA just once?
Walt B
Quote from sals3r0:
Did a big update today - you can create studies. Explanation:
Studies are essentially multi-backtests performed on whole ranges of indicator periods. If you create a study for pair of stocks, it will silently create backtests for all period ranges (10-50) at the same moment. Results are then displayed using a scatter plot of Compound Annual Growth Rate (CAGR) in one axis and maximal draw-down in the another. In addition, stock prices are displayed in a chart together with multiple correlation indicators to give further insight about pair history.
Studies are powerful tools used mainly to evaluate the robustness of the pair, how those stocks hold up for various settings. Because you can actually see the behavior of the pair for all possible parameters at once, this tool can be used to help eliminating the risk of over-optimization (i.e. some pair is bad, but for some particular periods gives good results - with a single backtest, you might not notice that). In general, good robust pair should exhibit satisfying results for all indicator periods in the test instead just for few.
Example: http://www.pairtradinglab.com/studies/TwnfcIhdDeO-DT3N
Quote from waltbx:
Looks interesting. Haven't got it figured out yet, but obviously a great project requiring a lot of work.
Is there a way to search for highly profitable pairs, or must we bring our possibilities to the tester to confirm suspicions on profitability?
(The CAPTCHAs are really hard to read).
Walt B
Hello, right now the website is just meant to be a tool to confirm suspiction/profitability just as you say. But there are plans to extend its functionality also to cover other uses in future.
Sorry about the captcha, but it needs to stay there for security reasons (the website would be immediately misused to screen stocks pairs without it). However, in the future we plan to introduce user registrations. Then captcha rules might be loosed for users who are not making troubles.
Quote from sals3r0:
Did a big update today - you can create studies. Explanation:
Studies are essentially multi-backtests performed on whole ranges of indicator periods. If you create a study for pair of stocks, it will silently create backtests for all period ranges (10-50) at the same moment. Results are then displayed using a scatter plot of Compound Annual Growth Rate (CAGR) in one axis and maximal draw-down in the another. In addition, stock prices are displayed in a chart together with multiple correlation indicators to give further insight about pair history.
Studies are powerful tools used mainly to evaluate the robustness of the pair, how those stocks hold up for various settings. Because you can actually see the behavior of the pair for all possible parameters at once, this tool can be used to help eliminating the risk of over-optimization (i.e. some pair is bad, but for some particular periods gives good results - with a single backtest, you might not notice that). In general, good robust pair should exhibit satisfying results for all indicator periods in the test instead just for few.
Example: http://www.pairtradinglab.com/studies/TwnfcIhdDeO-DT3N
I really really want to understand this. Can you explain more. I am obviously little slow for this discussion.
thanks
Quote from amitkumar_ny:
I really really want to understand this. Can you explain more. I am obviously little slow for this discussion.
thanks
OK Let me explain:
There are few approaches how to trade pairs. The most popular one (also the one we use at pairtradinglab) works with ratio between two stock prices together with two technical indicators applied on the ratio time series - moving average and standard deviation (you could also look at it as on the one indicator - bollinger bands, which is the same anyway). For these indicators you need parameters - period (how many days to look back).
So for pair trading strategy, you have at least two parameters:
1) moving average period (in days) - lets call it MP now
2) standard deviation period (in days) - lets call it SP now
In the end you have even more parameters (thresholds, ...) but I won't comment on them now, although everything said below applies on them too.
Then, in order to trade pairs successfully we have to make decisions:
1) which pairs to trade, how to judge pairs
2) what MP/SP combination to use a parameters to the strategy
Looks simple but actually this is a big magic.
About how to judge pair - we can do backtest and see if result is good enough - but then we still have problem: we need to select MP/SP also for this backtest! how to do it? we have quite a risk here - the whole pair may be crap, but just by luck, the backtest can be good for particular MP/SP tried...but if we were trading this pair in future, it would most likely turned to be a loser...because we built our assumption on something what was not really robust enough
How we can help ourselves with this?
We will not evaluate a pair just by a single backtest. Instead, we will perform backtests for large number of MP/SP combinations and observe, how results behave for those. We can then probably expect, that if the stock pair behaves well for ALL the parameter combinations instead of just one, we should have much more chance the pair will hold up in future trading, right?
So thats what basically study in pairtradinglab does.
It will perform backtests for all MPs in range 10..50 and SPs in range 10..50. So in total it will perform 1681 backtests, one for each period combination.
Then it will display for each result the CAGR and Max Drawdown of the backtest in the scatter plot, color determined by the period.
So in the scatter plot you can instantly see, how the stock pair behaves for ALL period parameters! You can instantly tell, if the pair is crap (it haves a lot of points in CAGR range below zero), or good - haves all points above zero in CAGR and good dd for all points.
So finally, the study feature helps solve both problems:
1) check if the stock pair is good/robust - good and robust pairs will indicate >=99% points above zero in CAGR and indicate good drawdown for them too
2) choose optimization parameters MP,SP - for instance, I would advise to take some big cluster of points having good CAGR and low DD and take the middle point of the cluster
I hope it is more clear now.