Arbitrage Strategy

Discussion in 'Strategy Development' started by Hammy27, Dec 15, 2011.

  1. Hammy27


    Hy there

    I do not really believe that (short to medium term) single stock strategy trading will be profitable in the long run (unless it's HFT). But I beliefe that you seriousy could be earn money with arbitrage.

    I therefore studied a few academic papers and applied/modified an existing strategy that (in my version) invests in a 1week timeframe. I tested the strategy and the statistics are the following:

    stats (monthly * 01/2005-11/2011):
    mean (ann.): 17.8%
    deviation (ann.): 12.4%
    skew: 3.8
    ex-kurtosis: 19.06
    #pos months: 57
    #neg months: 27
    best return: 23.16%
    worst return: -3.68%
    mean return (if positiv): 2.62%
    mean return (if negativ): -0.90%

    yearly returns:
    2005: +5.40%
    2006: + 5.00%
    2007: +9.00%
    2008: +46.6%
    2009: + 25.6%
    2010: +6.2%
    2011: + 24.8%

    My goal is not to earn money by investing my own money but investing others and earn a fee (maintenance&incentive).
    a) How long should my track record be to attract institutional money?
    b) If YOU would have this stats, would you give the strategy a "go ahead"?

    thanks for you replies and please apologize my English.
  2. Five positive years with little equity volatility and at least 50 Million under management to go after institutional money. Maybe now is even harder.

    Can you offer some details?

    1. How may trades/issues in you backtest
    2. Why didn't you test even further back?
    3. How many variables did you optimize?
  3. bone

    bone ET Sponsor

    You will not find a weekly arbitrage, when deep pockets and big money ECNs are chasing millisecond arbitrage for stupid size - and they are paying top dollar for the order flow.

    Your prints are likely deceptive artifacts - use live data for individual components and reconcile them for time & sales lag / bid-ask. Your picture of the arbitrage spread differential will change dramatically.
  4. Hammy27


    1. Trades? The Basket I trade is about 10 Stocks, so weekly repositioning = 10*52*5 = 2600 Trades. Sometimes you do not have to trade when the modell doesnt change weights.
    2. I did:
    2000: +18.55%
    2001: +14.58%
    2002: +25.24%
    2003: + 13.99%
    2004: -7.43%
    3. theres no optimization

    @bone: As I said, I trade the hole portfolio once a week. I don't care about bid/ask. I took closing prices (- 0.25% if short / +0.25% if long). I think this should compensate for spreads an slippage and so on...

    I'm now forward-testing the portfolio on bloomberg.
  5. So what are you arbitraging? This sounds like a typical position trading system based on some high levels filtering of stocks.

    Arbitrage means the simultaneous buy and sell of an asset in two different markets for taking advantage of mispricings. I don't see how you do that with weekly position trading. There is no arbitrage in weekly trading as Bone already said.

    Selecting 10 stocks to test out of 7,000 or so is the most serious optimization of all. How do you select those stocks? If you select them arbitrarily/randomly then you may have something. If you selected them so that your system works in testings, you have no edge.
  6. Hammy27


    Oh it's not arbitrage in the sense that I buy the same stock on different exchanges to generate a profit. As you said, I wouldn't have an edge!
    But I can do some stats on serial correlation, stationarity and so on. And cause I believe it's hard to compete on 2-stock basis I built a portfolio.

    How I choosed the stocks? I asked Bloomberg to display me the ones I need to generate the maximum possible profit... :)
    No seriously, I've got 2 criterias which I use for selection. I first choose a stock that I like and then I'll let Bloomberg find the 9 others I will need to build the strategy. So I do not fit anything. And: The stocks are the same for the whole time-frame.

    But back to my initial question: if we assume that the backtesting is accurate, would you invest in such a strategy?
    I'm currently not sure if I will... I will probably trade the portfolio on Bloomberg first for to see how it goes in the near future.
  7. To echo bone, your data is not synchronized and accordingly your backtest is useless.

    So, no.

    This also sounds like a basic lead/lag basket strategy, nothing remotely resembling arbitrage.
  8. Canadian


    a) probably longer than 0
    b) no, because they are not real
  9. Well, im wondering how do you present an arb data then, since every trade you take = locked in profits.

    I have done only a few arbs in my life because I don't monitor them, but it definitely feels good to spot one and get the MM (the party the other side) to take your orders.

    Since most arbs (what I hv done) can only make say a 0.2% cut (risk free though) is that even significant ? btw these arbs cannot high amoutn of $, else it just goes away. So for eg. I can only put in $5,000 and make a $100, but if I put in $50,000 I may make only $300 or less.

    Anyone any ideas ?
  10. Pair trading stocks is not arbitrage. Correlations can change. Some pairs take forever to converge = tie up capital.

    Statistical analysis is only as good as your data. How many years of data did you use? Did you account for survivorship bias? Backtesting will never expose to the risk of black swans.

    Hypothetical example: Long KO/Short PEP... You learn PEP Is cooking it's books... KO remains $70, PEP goes to $1... what do you do?
    #10     Dec 26, 2011