Author Topic: For those who are mathematically challenged...  (Read 3284 times)

Paul Jessup

  • Newbie
  • *
  • Posts: 4
  • Karma: +0/-0
For those who are mathematically challenged...
« on: January 11, 2015, 03:25:34 pm »
What I am looking for are the pairs with the best chance of keeping moving in opposite directions, say to 2 standard deviations (and currently at or near that level),  returning to the norm, and moving 2 standard deviations in the other direction, and back to the norm, and so on. ES vs. YM, miners against gold, bonds against stocks, March against April, etc. I would strongly suggest adding in commodities and the main Forex pairs.

Standard deviation would seem the obvious candidate to pick stocks, but I suppose your co-integration, and perhaps other filters, if I knew what they meant, and knew whether the figure should be higher or lower to get to my goal!! Why not have set filters which suggest different strategies?

Then I would like to look at them using your chart, pick say the best 10 pairs, and have them entered into the program. It would be good if the web page told you how many stocks it had found with your search pattern, without first putting them on the screen. A bit like mobile.de.

The problem is - I am faced with so many mathematical formulae, I have no idea which are the best! I imagine that different trading houses have different strategies, but the retail chap like me - well, we want to know:

1. Have the stocks high implied volatility at present?
2. Have they liquidity?
3. What is the maximum profit one could expect (and then aim for say 40% of that)?
4. What's the probable time frame of getting to maximum profit?
5. What's the likely-hood of getting there?
6. How do we set the filters to achieve that?

We need to keep seeing it on a chart too. Of course, I could use IB for that, so it's not a first priority. But your Delta chart looks very interesting. Similar to the spread chart I see when trading HO against RB, or NG March against NG April. That's probably the best way to explain this. What I am looking for are 10 stock pairs with a Delta graph being as near to a smooth sine wave both sides of 0 as possible, and to enter the trade at the furthest point from 0!!! And get out the other side of course!

admin

  • Administrator
  • Sr. Member
  • *****
  • Posts: 320
  • Karma: +12/-0
    • Pair Trading Lab
Re: For those who are mathematically challenged...
« Reply #1 on: January 11, 2015, 05:01:51 pm »
Quote
What I am looking for are the pairs with the best chance of keeping moving in opposite directions, say to 2 standard deviations (and currently at or near that level),  returning to the norm, and moving 2 standard deviations in the other direction, and back to the norm, and so on. ES vs. YM, miners against gold, bonds against stocks, March against April, etc. I would strongly suggest adding in commodities and the main Forex pairs.

Yes, everyone is looking for pairs like this. In other words you are looking for pairs which have their spread or ratio reverting to the mean as often as possible without having too many extremes.

At this moment, we don't have plans of adding Forex market support. There are not much markets anyway and they are all heavily manipulated by big players (mostly banks). Personally I don't know any successful pair traders operating on Forex markets. All successful pair traders I personally know operate on stock/ETF markets.

We might add futures market support in the future, but it is not a priority either. There are so many more markets with greater potential (like non-US stock markets).

Quote
Standard deviation would seem the obvious candidate to pick stocks, but I suppose your co-integration, and perhaps other filters, if I knew what they meant, and knew whether the figure should be higher or lower to get to my goal!! Why not have set filters which suggest different strategies?

I would strongly suggest to start studying basic pairs trading principles first (correlation, cointegration, mean reversion, ...).  As a good start I would recommend to read this book:

http://www.amazon.com/dp/1118460146/ref=as_sl_pd_tf_lc?tag=quantitativet-20&camp=14573&creative=327641&linkCode=as1&creativeASIN=1118460146&adid=1CWM7P466WXGVFFYQFQF

(especially the "mean reversion" section)

Unfortunately we don't provide any tutorials or lectures at this moment, but in near future we want to create some tutorials. We will also provide (paid) consultations over e-mail or Skype for clients what are not sure about their approach to pairs trading. We will also have a blog site with interesting articles and tutorials.

Quote
Then I would like to look at them using your chart, pick say the best 10 pairs, and have them entered into the program. It would be good if the web page told you how many stocks it had found with your search pattern, without first putting them on the screen. A bit like mobile.de.

Good start is this: use our Database feature - filter for pairs which have:
  • medium CAGR at least 20 for last 5 years
  • min CAGR at least zero for last 5 years
  • medium cagr at least 0 for last 6 months
  • average volume at least 100 000
  • and sort your results using Median Equity Linearity in a descending order

You will find many interesting pair candidates already using this.

Right now we are working on even much more powerful pair database system having millions of pairs.
Quote
The problem is - I am faced with so many mathematical formulae, I have no idea which are the best!
I know, pairs trading is not trivial and there are many approaches. Basically, you need to select your pairs using backtests and metrics like cointegration based on historical data. Then you need to do out-of-sample tests and trade your portfolio in paper mode for the while. For pairs trading apply same rules like for any algorithmic trading - testing robustness, walk-forward backtesting, etc...

We will also provide full-featured online genetic optimizer with robustness checking, which will answer your question about what model and parameters give best results.

To your questions:
Quote
Have the stocks high implied volatility at present?
Which stocks do you mean? Whole stock market in general? You can look at the VIX index and look for yourself.
Quote
Have they liquidity?
You choose stocks with acceptable liquidity yourself. Good lead here is to check average daily volume.
Quote
What is the maximum profit one could expect (and then aim for say 40% of that)?
I think good pair traders can do between 20-40% a year. I know some traders achieving this personally. They trade stocks and ETFs.
Quote
What's the probable time frame of getting to maximum profit?
Quote
What's the likely-hood of getting there?
Quote
How do we set the filters to achieve that?
I am sorry, I don't really understand these questions.

Quote
We need to keep seeing it on a chart too. Of course, I could use IB for that, so it's not a first priority. But your Delta chart looks very interesting. Similar to the spread chart I see when trading HO against RB, or NG March against NG April. That's probably the best way to explain this. What I am looking for are 10 stock pairs with a Delta graph being as near to a smooth sine wave both sides of 0 as possible, and to enter the trade at the furthest point from 0!!! And get out the other side of course!

I would not spend too much time looking at Delta chart as it can be misleading. Look at cointegration and look at backtests.
You want pairs with sufficient performance in backtests with equity as linear as possible. That is my start advice. You can already use the database to achieve this.
We are now working on the new database system which will also include cointegration.

Paul Jessup

  • Newbie
  • *
  • Posts: 4
  • Karma: +0/-0
Re: For those who are mathematically challenged...
« Reply #2 on: February 03, 2015, 01:45:02 pm »
Thanks for the help. I am aware of mean reversion, and its use in trading. However, standard deviation may show how far the pair move away from parity. Volatility plays a part here.

Perhaps FX doesn't work, but commodities certainly do. I trade HO/RB for example, selling when the pair get to or near parity. But I do that without need to search a database, as the pairs are limited, or I use the spread between different months for the same underlying. Anyway, as I said, just a suggestion.

I have followed your suggestion of the various pairs settings, and I will look at the CSV files. Thank you for the guidance.

Could you please provide a link for the on-line optimiser. I couldn't readily find it.

IV Rank - stocks with earnings coming up are usually more volatile than those which don't. For example, when earnings have past, volatility collapses. This is the volatility of the pair, and can be calculated from the options. I was just thinking that PLT would be an even more complete product. The VIX and its siblings, are of no interest.

What's the probable time frame of getting to maximum profit?, What's the likely-hood of getting there?, How do we set the filters to achieve that?

I am asking here if any of your filters would show how quickly one could expect the difference in the stock prices should return to the norm, and how likely that is.

I like an artistic representation of what is going on in the trade. It's not a must.

It seems that some of your planned improvements would answer some of my questions. Keep up the good work!

admin

  • Administrator
  • Sr. Member
  • *****
  • Posts: 320
  • Karma: +12/-0
    • Pair Trading Lab
Re: For those who are mathematically challenged...
« Reply #3 on: February 05, 2015, 09:55:13 am »
Quote
Could you please provide a link for the on-line optimiser. I couldn't readily find it.

Sorry, there is no online optimizer yet. But we would like to have it done and available later this year.

Quote
I am asking here if any of your filters would show how quickly one could expect the difference in the stock prices should return to the norm, and how likely that is.

You can take the half-life estimate for this. The lower, the better (mean reversion coming more often). You can already see this when you analyze a pair (using the online analyzer tool). But the pair should be cointegrated in order this assumption works.

The new database product we are working on now will include this metrics for searching.