Questions:

• The ML based AiLA allocations originate from identified opportunities, given the prevailing market conditions.
• The percentage of allocated capital can therefore vary significantly, as well as rapidly, over time.
• Can we force the index to maintain its allocation percentage within a given range?
• What is the corresponding impact on performance?

Approach:

• Illustrate impact for an AiLA index based on 102 selected assets using long-only (LO) allocations, for which allocation tends to vary more over time.
• Implement allocation range functionality based on adjusting holding periods associated with active signals and awareness of contract maturities.
• Assess magnitude of performance impact from such allocation range criteria.

• The allocation percentage is affected by different constraints and targets in the index construction process.
• However, often the number of times of allocation is the main aspect for how much capital that is allocated, which in turn depends on the ML model’s ability to identify opportunities.
• This results in periods with few trade opportunities and hence low allocation, which change as conditions vary.
• For the comparison, the allocation percentage is forced to be sustained within a given (min/max) range by,
  • Forcing allocations to be applied outside the tenor set that is considered optimal by the system.
  • Adjusting holding periods when necessary.
  • Prioritize w.r.t maturities among contracts with signal.
• The methodology is not expected in any way to be optimal, but intended to serve as a first order assessment of the impact from this type of functionality.

• The index performance was investigated for different allocation percentage ranges, i.e. within a min/max value.
• Forcing the index to maintain a minimum allocation percentage reduced the performance noticeably, e.g. by requiring an allocation percentage in a range close to 100% reduced the performance by about half.
• AiLA index performance appears on a risk adjusted basis,
  • Higher than from a standard index like GSCI/BCOM.
  • However, the correlation to GSCI/BCOM increases with allocation range converging towards 100%.
  • Typically, the correlation increase from about 60% (AiLA LO) to almost 80% (100/100).
• Sustaining a larger allocation by forcing the system away from its specific trading recommendations, significantly reduce the performance due to the opportunistic nature of the ML based AiLA signals.