Fully automated futures trading

Discussion in 'Journals' started by globalarbtrader, Feb 11, 2015.

  1. maciejz


    Based on reading the blog post, I think I understand the approach of pooling gross returns but using instrument specific costs. My understanding is that we pool the gross return data of all instruments in our group (typically asset class). When performing the bootstrap optimization, we use the pooled set of returns, which would lead to same weights for all instruments. But, we perform the bootstrap optimization for each instrument separately and we use the costs for each instrument to "adjust" the returns from the pool during the optimization. The end result should be a set of weights that are different for each instrument based on the difference in instrument trading costs. My understanding is that you also remove any rules that exceed the speed limit prior to the optimization.

    Yes, I get the index data for the S&P500 and STOXX to calculate the historical volatility; I use the VIX and VSTOXX index (as opposed to futures) values as the implied volatility.

    For the carry curvature, I loosely follow http://www.cboe.com/rmc/2015/Day-1-Session-1A-Berlinda-Liu.pdf. I calculate the DISTANT_TERM_CARRY as the average of carry between M4 and M5, M5 and M6, and M6 and M7. The reason is that there is seasonality (December) in the VIX term structure and I'm hoping the averaging of the several carry calculations removes the seasonality. I also calculate a NEAR_TERM_CARRY as the average of the M1:M2, M2:M3 pairs. Then, CARRY_CURVATURE = NEAR_TERM_CARRY - DISTANT_TERM_CARY.

    I'm not sure the Curvature approach would be very useful for other instruments since my understanding is that the only other instruments that have any volume in delivery months beyond the front month also have a lot of seasonality. Perhaps EuroDollar is the exception though.

    Thanks for the pointers.

    Last edited: Sep 15, 2017
    #991     Sep 15, 2017
    globalarbtrader likes this.
  2. isotope1


    What are your thoughts on dynamic allocation to trend following?

    I'm following the lead from the last chapter of the Kaminski book. I'm going to swap the equity part of my portfolio to a Vanguard 60/40 fund, but the book proposes capitalising on the mean reverting nature of TF's performance, and shifting the allocation as such.

    Obviously I know only a backtest might answer this question (not sure if there's enough data to say, I will try to incorporate CGT into the backtest), but just curious if it's something you'd explored.

    PS If the answer's in the new book, I'm looking forward to reading it! :)
    #992     Sep 20, 2017
  3. Well the answer isn't in the new book, although I discuss changing weights given some information about future SR in table 12 of the first book.

    But I can tell you that whenever I've tested this in the past it's been difficult to find a statistically significant effect. Bear in mind you'll be increasing your trading costs as well. Also you'll be ramping up your leverage in periods when TF has done badly - make sure you'd be happy doing that.

    If you insist on doing this keep the shift small; i.e. don't change your allocation if it's 50:50 beyond 60:40 (as table 12 suggests). Of course this means the benefits will be small as well - again begging the question of why bother.

    #993     Sep 20, 2017
  4. As most of you know I run two types of strategy in my trading account: (a) a long ETF portfolio of about 80:20 stocks / bonds with a very imperfect hedge (which consists of a short position in Eurostoxx), (b) a futures trading strategy. As I've discussed before I didn't want to fully fund my account with cash as it would mean incurring CGT. But I also didn't want the long only ETFs to 'pollute' my return stream. So I did the hedge.

    Anyway I've decided to close the hedge. The main annoying that is that the hedge gives my account a real long EURGBP bias, which although I don't have an opinion on has been irritating with all the swings in that currency over the last year or so. Economically this is the right thing to do since in the long term the hedge should cost money (I have no opinion on the short term direction of eurostoxx). I'm decided I am not really that bothered if the account is a pure futures account or not. I can still easily attribute my p&l to each component (in fact it will be easier).

    Because of the way my system works (daily trading limits) it will take 3 days to close the position of 9 contracts.

    First three contracts done today at 3509.0 versus the close of 3513.0 :) and £40 of slippage :-(

    Clearly everyone should now go short Eurostoxx as this is bound to be the wrong decision!

    #994     Sep 20, 2017
    JMW likes this.
  5. As an update, all the closing trades are now done. The average price was 3517.67, with £67 of slippage.


    #995     Sep 22, 2017
  6. Elder


    Hi GAT, I'll be following with some interest the performance of your evolved of your portfolio, I guess it makes sense as it is a little bit more diversified in terms of strategies (long + systematic)?

    Another quick question for you on the systematic side: I note that you use closing prices for your daily data. Quandl seem to mostly publish settlement prices for most ref futures. Is there any reason why using settlement prices would be inferior for backtests and signal generation in comparison to using closing prices?
    #996     Sep 26, 2017
  7. It shouldn't make much difference, unless you're trading spreads where you'd need to make sure whatever you were capturing was a tradeable price.

    #997     Sep 26, 2017
  8. joederp


    Is this a function of the 'looseness' of your trading rules? I.e. some EWMAC of EUR/GBP?
    #998     Sep 26, 2017
  9. I'm not actually trading EUR/GBP, just had an implicit exposure through my hedge.

    #999     Sep 27, 2017
  10. Elder


    Thanks GAT, I am assuming calendar spreads for carry calcs are no less tradeable based off settlement prices rather than closing prices. Intermarket spreads I will leave well alone for now. The only slightly weird element would be the need to timestamp settlement prices as end of day when in fact they might occur in between certain hourly snapshots.

    BTW good luck in Singapore, will try to make the next one!
    #1000     Sep 27, 2017