Intraday Pair Trade

Discussion in 'Trading' started by ar1zona, Mar 29, 2011.

  1. ar1zona



    Is there anyone here still trading equity/ETF pairs on an intraday basis? Searched some of the older threads found a few members who used to trade this way and claim to be successful.

    I just made an excel sheet backtesting using 1-min and 5-min bars and the results were really bad. Double checked with spread charts all the entry and exit signals checks out so its not a problem with the sheet.

    Another question is what kind spread do you use? I know for swing pairs plotting (price A / price B) works fine but some people seems to use (price A - price B) as well. I even used my sheet to test some of the pairs that other intraday pair traders used (i.e. IWM-SPY, along with some stock pairs) but it still shows a net loss. Would really appreciate any experienced traders to point to me a direction or give some advice. I am willing to share my backtesting results and methods so we can cross-check what's wrong with my strategy. Thank you in advance for any help.
  2. Whats the correlation percentage between your pairs? I've found that I need less layers (smaller SD), when the percentage is over 85% over the last 12 months.

  3. ar1zona


    I kind of tried both high and medium correlations.

    For equities I backtested, I have a sheet from someone else who used to do pairs intraday. He had 96 pairs and that was half a year ago. I retested those pairs but only a few turned out to be profitable. Even those that are profitable, the result is only valid for a few months. This is no consistent relationship for the two stocks to cointegrate.

    For ETFs I used Market Topology. They have a list of highly correlated ETFs. Surprisingly I tested like 10-20 and found none of them to work out. Every single pair resulted in losses over a 2-month period after paying the spread.

    Seriously dunno what's wrong with my sheet so I will briefly explain how I made the sheet and see if anyone can point out an error or something.

    1. import 1-min or 5-min bars for two stocks, import 2-months worth of bars or whatever time frame, double check if the time and date all matches
    2. divide stock A by stock B for each bar (used closing price), calculated bollinger band based on varies combination of lookback period and standard deviation (i.e. 100, 2)
    3. when the ratio penetrate the bollinger band twice (first time diverging, second time converging back to the band), take trade
    4. get out at market close or when ratio hits moving average

    Again, thanks in advance for anyone willing to help. Intraday pair may not be as good as swing pairs but I know there are people doing it so there has to be a way. I just need someone to point me the right direction...
  4. I don’t trade this way myself, but suspect that if this opportunity (intraday pairs trading) does exist it won’t by now be accessible to the retail trader.

    “Every” algorithmic trading and HFT firm has its own statistical arbitrage model (I believe “pairs trading” falls into this category) and in order to compete will have invested time and money in setting up execution capabilities a retail trader won’t generally have access to. By now, if this opportunity exists, it will be the fastest, best executing of these players that gets the trade, not a retail trader.

    If you do want to investigate further though, I would look in more detail at the bar data you are importing. Where does it come from? Is it from just one exchange (probably what you need for this strategy, so that you can consider collocating close to the data source of this exchange), or is it some undefined composite of data from lots of exchanges (probably useless for this strategy, as it won’t reveal local actionable opportunities where mispricings do occur)?
  5. You need TAQ data to watch intraday pairs. 1- and 5-min bars won't cut it in this environment.
  6. You can definitely make money intraday trading pairs. A few things to consider:

    1. Dollar Neutral is not the best ratio for pairs trading
    2. Wait for the move to start back in rather than catching a falling knife
    3. Focus on sectors that have the vol right now.

    Long term pairs is a lot more tricky. Earnings, buybacks, secondaries, dividends, and other corporate actions all come into play. But those 3 things above still matter.
  7. intraday doesn't suit gotta swing it..........
  8. Bob111


    what is the difference between pair and directional trading anyway?
    i see no difference. you still trading the trend.and yes, i done intraday pairs and done it well.
  9. bone

    bone ET Sponsor

    I have taken on a couple clients who used to trade pairs for Bright at the Vegas and Vancouver offices, and are currently making a transition into futures spreading. My perspective:

    1. Very tough to intraday equity pairs trade for convergence; i.e., fading moves according to a 'rich - cheap' strategy.

    2. Baskets work better than pairs IMO if you are looking to hold a move without subjecting yourself to turbulence, and baskets make for better swing-trading opportunities IMO. The baskets will most definitely model and chart smoother - they 'behave'.

    3. If you are automating a strategy, you should most likely use a different exit rule than the entry rule you utilized. That is a very big deal that's frequently overlooked.
  10. ar1zona


    Sorry I forgot to mention this I have prop firm account so there's access to real time tick data. And although I did backtesting with 1-5 min bars (from Qlink) the execution is intended to be reading tick data real time from Sterling T&S.

    The main reason I used 1- and 5- min bars is because tick data have too many noise. And its impossible to have ticks for both stocks at each specific second. Using bar closing price at least there's a consistency in data. I actually do find 5 min worked a bit better than 1 min, and heard people talking about even 1-5 min bars have too much noise and its better to use 10-20 mins if you have to trade intraday. Do you think its worth to test with a smaller timeframe (subminute) and what kind data should I get for that? Thank you.
    #10     Mar 30, 2011