CML Complements AML

Questions

Questions:

  • The CML signal performance appears relevant given the enhancement indicated wrt traditional CTA signals (see CTA modelling section).
  • However, are the CML signals relevant for a limited number of futures markets?
  • Also, are the CML signals complementary wrt the other AiLA signals?

Approach:

  • Start from index based on AiLA CML long-only signals for 9 futures markets (CML-LO), i.e. an index intended to have 𝜌 ~ 30 - 40% to GSCI and BCOM.
  • Illustrate implications on the index from capacity limitations and when that becomes relevant.
  • Investigate performance overlap between the CML and other AiLA signals to assess complementary value.

CML stands for CTA Strategies enhanced by Machine Learning
AML stands for AiLA Strategies developed using Machine Learning

Capacity Limitation

  • High index capacity is very challenging when based on a limited market set, with large difference in liquidity.
  • Here we use average volume within one month from planned roll date, e.g. about 60’000 lots from RBOB gasoline over the last 10 years, translated into notional value using an average price over last 6 months.
  • The ability to only rebalance index weights according to a small fraction of daily volume limits the possible market weights, and in particular for the least liquid markets since volume differ by more than one order of magnitude.
  • For a large allocation to the CML long-only index, e.g. 500M USD, the low number of active signals per day together with liquidity only allowing a small fraction to be allocated to each market, will limit the index capacity.
  • Therefore, complementary market signals will be important also in order to share allocation capacity.

(* The returns used are based on the front month contract for each market, typically rolled a couple of weeks before the contract expiry)

Complementary Signals

  • The CML signals are compared to other AiLA signals using the same 9 futures markets and long-only signals (AML-LO).
  • These signal/market similarities are expected to results in positive performance correlation, however, relatively low values were observed, i.e. typically ρ< 40%.
  • The low performance overlap primarily originates from the CML signals identifying opportunities, i.e. allocating weights, at different times than the other AiLA signals for a given futures market.
  • This complementary nature between CML and the other signals is expected to be beneficial in terms of,
    • Performance, e.g. by identifying complementary opportunities over time,
    • Allocation capacity, e.g. by a larger number of allocations to different markets at a given time.

(* The results here are not including any liquidity constraints, in favor of a cleaner comparison with respect to the signal overlap)

Conclusions

  • An index was investigated based on long-only CML signals applied to only 9 futures markets with the intention of having a long commodity market exposure, e.g. with a positive index correlation of about 30 - 40% to GSCI or BCOM.
  • Despite the small number of markets, the index performance appears relevant compared to common strategies, such as trend-following which often require a much larger number of diversifying markets, however, with limited index capacity.
  • In a signal comparison using the same markets and long-only configuration, the CML signals indicate a complementary nature with respect to other AiLA signals in terms of identifying opportunities, i.e. allocating index weights, at different times for a given market.
  • This indicates complementary value of the CML signals to that from the other AiLA signals, both in terms of index performance, i.e. from additional un-correlated returns, as well as increased allocation capacity, i.e. from additional market weights at a given time.