Stat Arb: Pair Trading (US Equity)

Discussion in 'Strategy Development' started by Esq Esq Esq, Mar 23, 2009.

  1. Hey everyone,

    my team does a shitload of pair trading and we are facing some minor liquidity issues (since we trade a lot of size and very active).

    So we got together and were thinking about creating a spread-exchange / dark pool for spreads.

    We all know the hurdles are the bid/asked spread on each stock + remove liquidity fees. Those sometimes may eat away at the spread's movement.

    However, since many pair traders have different time horizons and different entry/exits, there may be some sense for them (us) to trade with each-other to minimize bid/asked spread , remove fees and overall slippage.

    We have a big development team, so there isn't an issue as far as making this a reality from a tech standpoint, but want to see if there are others that feel that they could benefit from this endeavor.

    If there are existing spread pools already - please please post. The more liquidity the better.

    Best of luck trading to all.
     
  2. Thats an interesting idea however if you think it thru it is somewhat of self-contradictory :)

    The vast majority of pair-arbs, pretty much does the same things. All model it in diff ways, all implement in diff ways however...the idea is that *most* entries for a spread would evolve around a certain point and so would most exists :) in essence, what Im saying is, there is decisive asymmetry between buyers and sellers for a given moment in time..i.e. there are very few cases where I would initiate a spread and you would be unwinding yours.

    However, if such pool, could do efficient reroutes at sustainable fees then of course it could really solve a *lot* of problems for a *lot* of people.

    from my other posts you can gain some more clue about my experience, but in essence, if instead of firing hundreds of orders at dozens of gateways trying to get my desired fill I could send ONE order to your (ours :D) pool and let IT do the matching and appropriate reroutes for me it could really make an impact on profitability. In that sense, the project could really be of great interest as it is both profitable in itself and could marginally improve conditions for its users.

    My regards :D
     

  3. Well - i did give it some thought.
    The really interesting thing, is that everyone trades in different time frames, with different entry parameters.

    Someone may want to scalp 1 - 1.5 sigma - another 2 - 3, another 3 -5, another 5 - 8 where as the first guys' out would be there

    Some bet on reversion, some bet on breakout (i do both).

    So, given enough different parties, there should be enough liqudiity at different price points in different ways.
     
  4. marine

    marine

    Interesting idea but probably unlikely. I would try to find a way to arb it.

    Another approach is to try to MM those securities. If you can find away to incorporate that into your model that would help significantly. Plus increase the amount of times a security appears in a pair. Then implement an internal dark liquidity pool. We do something similiar to this and it work nicely.

    A question for you:

    How many pairs does your team actively trade and how many traders? We (2 guys) actively trade 300 pairs. We are planning on trading 1000 pairs between us. From the little feedback I have received from asking this question on different boards everyone seems to think this is alot.
     
  5. I do about 100 pairs...
    But have a lot of great analytical infrastructure...
    So 300 by 2 guys would be comparable.

    The bottlenecks are:

    (1) Finding opportunities by scanning a universe.

    (2) Making a decision to trade.

    (3) Entering an order.

    (4) Real time administration.

    It depends almost entirely on your level of Automation...
    Which can range from superb organization with minimal automation...
    To adding simple auto-orders for no brainer scalping...
    All the way to very sophisticated Automation where 80-90% of your orders are auto entered...
    Always assuming that the entire system is monitored continuously by Pro Traders.

    Fairly soon you reach the point where you cannot trade if systems go down...
    So full time engineering staff is mandatory.

    Whatever you are doing... you could easily double or triple volume.
     
  6. This is real funny to me. Comical even. Pairs trades are daily events, not intraday time frames. You shouldn't have to worry about spreads or bid/ask or the frequency of your trades when you're pair trading.
     
  7. lol
    Look at SPY:IVV, IAU:GLD, ... there is no pairs trading intraday?
     
  8. ronblack

    ronblack

    To say the least...Pairs trades are for those who love double hits...
     
  9. nitro

    nitro

    Almost true, but 100% true if your definition of "pairs" only means stat arb.
     
  10. The Problem comes with them being .98 "POSITIVELY" correlated. For pairs trading to work, you need nearly perfect "NEGATIVE" correlation. You shouldn't expect any security to be under or overvalued relative to the other security when they both move in the same direction. My pairs model still works on these securities, but it's not very robust at the 30 minute level.

    In fact at the 30 minute level, my pairs model on IVV and SPY without modifications has these results:

    Long + Short Long Only Short Only Buy & Hold
    Starting Capital $57,000.00 $57,000.00 $57,000.00 $57,000.00
    Ending Capital $57,245.19 $57,245.19 $57,000.00 $52,161.01
    Net Profit $245.19 $245.19 $0.00 $-4,838.99
    Net Profit % 0.43% 0.43% 0.00% -8.49%
    Annualized Gain % 0.08% 0.08% 0.00% -1.62%
    Exposure 0.72% 0.72% 0.00% 100.00%

    Number of Trades 26 26 0 1
    Avg Profit/Loss $9.43 $9.43 $0.00 $-4,838.99
    Avg Profit/Loss % 0.02% 0.02% 0.00% -8.49%
    Avg Bars Held 5.50 5.50 0.00 19,050.00

    Winning Trades 15 15 0 0
    Winning % 57.69% 57.69% N/A 0.00%
    Gross Profit $1,125.65 $1,125.65 $0.00 $0.00
    Avg Profit $75.04 $75.04 $0.00 $0.00
    Avg Profit % 0.14% 0.14% 0.00% 0.00%
    Avg Bars Held 5.73 5.73 0.00 0.00
    Max Consecutive 4 4 0 N/A

    Losing Trades 11 11 0 1
    Losing % 42.31% 42.31% N/A 100.00%
    Gross Loss $-880.46 $-880.46 $0.00 $-4,838.99
    Avg Loss $-80.04 $-80.04 $0.00 $-4,838.99
    Avg Loss % -0.14% -0.14% 0.00% -8.49%
    Avg Bars Held 5.18 5.18 0.00 19,050.00
    Max Consecutive 4 4 0 N/A

    Max Drawdown $-904.23 $-904.23 $0.00 $-44,788.80
    Max Drawdown % -1.57% -1.57% 0.00% -57.27%
    Max Drawdown Date 6/29/2007 6/29/2007 N/A 3/6/2009

    Wealth-Lab Score 10.85 10.85 0.00 -2.55
    Profit Factor 1.28 1.28 0.00 0.00
    Recovery Factor 0.27 0.27 N/A 0.11
    Payoff Ratio 0.95 0.95 0.00 0.00
    Sharpe Ratio 0.18 0.18 0.00 -0.03
    Ulcer Index 0.26 0.26 0.00 19.29
    Wealth-Lab Error Term 0.15 0.15 0.00 12.50
    Wealth-Lab Reward Ratio 0.52 0.52 N/A -0.13
    Luck Coefficient 3.50 3.50 0.00 0.00
    Pessimistic Rate of Return 0.74 0.74 0.00 0.00
    Equity Drop Ratio 0.00 0.00 0.00 -3.93


    Not that robust, but at the daily level, they have these results:
    Long + Short Long Only Short Only Buy & Hold
    Starting Capital $57,000.00 $57,000.00 $57,000.00 $57,000.00
    Ending Capital $72,087.08 $72,087.08 $57,000.00 $51,982.40
    Net Profit $15,087.08 $15,087.08 $0.00 $-5,017.60
    Net Profit % 26.47% 26.47% 0.00% -8.80%
    Annualized Gain % 4.43% 4.43% 0.00% -1.69%
    Exposure 4.14% 4.14% 0.00% 100.00%

    Number of Trades 25 25 0 1
    Avg Profit/Loss $603.48 $603.48 $0.00 $-5,017.60
    Avg Profit/Loss % 1.00% 1.00% 0.00% -8.78%
    Avg Bars Held 2.36 2.36 0.00 1,364.00

    Winning Trades 22 22 0 0
    Winning % 88.00% 88.00% N/A 0.00%
    Gross Profit $19,628.92 $19,628.92 $0.00 $0.00
    Avg Profit $892.22 $892.22 $0.00 $0.00
    Avg Profit % 1.47% 1.47% 0.00% 0.00%
    Avg Bars Held 1.86 1.86 0.00 0.00
    Max Consecutive 15 15 0 N/A

    Losing Trades 3 3 0 1
    Losing % 12.00% 12.00% N/A 100.00%
    Gross Loss $-4,541.84 $-4,541.84 $0.00 $-5,017.60
    Avg Loss $-1,513.95 $-1,513.95 $0.00 $-5,017.60
    Avg Loss % -2.45% -2.45% 0.00% -8.78%
    Avg Bars Held 6.00 6.00 0.00 1,364.00
    Max Consecutive 1 1 0 N/A

    Max Drawdown $-1,976.70 $-1,976.70 $0.00 $-43,945.59
    Max Drawdown % -2.95% -2.95% 0.00% -56.62%
    Max Drawdown Date 7/8/2008 7/8/2008 N/A 3/9/2009

    Wealth-Lab Score 103.79 103.79 0.00 -2.64
    Profit Factor 4.32 4.32 0.00 0.00
    Recovery Factor 7.63 7.63 N/A 0.11
    Payoff Ratio 0.60 0.60 0.00 0.00
    Sharpe Ratio 1.20 1.20 0.00 -0.04
    Ulcer Index 1.06 1.06 0.00 19.14
    Wealth-Lab Error Term 1.59 1.59 0.00 12.54
    Wealth-Lab Reward Ratio 2.79 2.79 N/A -0.13
    Luck Coefficient 2.36 2.36 0.00 0.00
    Pessimistic Rate of Return 2.19 2.19 0.00 0.00
    Equity Drop Ratio 0.00 0.00 0.00 -3.35


    Not bad. Good PF, marginally high Sharpe, good win percentage, but with a statistically insignificant amount of trades. However, given that I haven't modified the program I normally use for QID and QLD, good nonetheless. My advice still stands for this person. Use daily, not intraday in your strategy. You shouldn't have to worry about bid/ask spreads or commissions when pair trading.

    The problem with the time frame comes from the fact that you keep crossing under and crossing back over the fair value range. I would agree at the small level, there's not enough entry into an overvalued range to give enough of a profit margin intraday, thus why you must look at daily time frames.

    To summarize, the main problem with intraday time frames is that as soon as you cross that magic level of oversold or overbought, you buy, no matter how insignificant that discount from fair value is. So if you are looking at a 5 minute chart, and it goes below your fair value by a penny, then you put in the trade on for that penny, which I can then understand why there would be issues of trading costs. But if you are at the daily level, and it goes below the fair value by a couple percent, THEN, you have a much more robust system.

    The thing I've always noted with the look of the ratios in positively correlated pais, is that they move too sharply, and you need a negatively correlated pair to "smooth" the series out so that when one goes down, the other goes up, and you lose the choppiness of the ratio. This advice is quite valuable, but, of course, assumes the OP knows what to do with it.
     
    #10     Sep 12, 2009