Methodology Matters

Strategy Construction Methodology

4 Pillars of Methodology

  1. Signal Generation
    • Primary allocation signals generated by machine learning (ML) models trained using factor data, for opportunistic allocation
    • Clear delineation of the training, validation and out-sample periods to eliminate overfitting
  2. Risk Optimization
    • Aligns allocation to underlying market price cycles
    • Resultant risk reward of at-least 1:3 with a maximum holding period of 30 days, provides guard rails at point of allocation
  3. Allocation Caps
    • Weights allocations to assets are made to respect overall Asset, Sector and Group cap constraints
    • Rebalancing caps applied align final allocations to market liquidity, making the strategy executable
  4. Portfolio Construction
    • Each strategy is based on Markowitz model and built by including liquid future contracts
    • Independent return streams are reinforced by applying diversification, and allocation across the curve
In summary, AiLA follows a differentiated strategy construction approach to alpha generation with uncorrelated returns​

Overview

Background

AILA strategies can be classified into three different types:

  1. Seasonal Strategies
  2. Event Driven Strategies
  3. Trade Flow Strategies

These strategies were then decomposed to see if there is:

  • A strong correlation to a particular factor
  • Can any single or set of factors explain most of the alpha
  • Is the alpha truly a residual which cannot be explained by these 14 factors

Strategies Considered for Analysis

  1. AILAS002: Euro power & gas strategy
  2. AILAF010: Diversified commodity strategy
  3. AILAF012: Event driven diversified Commodity
  4. AILAS043: Agriculture strategy - Grains, Oilseeds and Softs
  5. AILAS066: Base metals strategy

Regression Methodology

Multivariable regression and correlation analysis of return streams of each strategy with factor returns streams:

Regression Formula

A total of 14 factors were considered in the evaluation:

Standard Factors
  1. Congestion
  2. Curve Carry
  3. Backwardation
  4. Trend
  5. Value
  6. Vol Carry
  7. Commodities Beta
Sectoral Factors
  1. Base Metals
  2. Precious Metals
  3. Oil
  4. Grains and Oilseeds
  5. Soft Agricultural
  6. Gas
  7. EU Power

Contribution

In the following analysis, each strategy is assessed against the 14 factors in consideration. The statistically significant factors are highlighted, and the residual is considered as unexplained alpha.

None of the strategies are correlated to the 14 factors chosen here.

AILAS002

EU Power Seasonal Strategy

  • The strategy is uncorrelated to all the 14 factors considered and demonstrates it has a residual alpha of 86.4% which is not explained by the 14 factors
  • The key factors which have a statistically significance are Backwardation and EU Power Beta
AILAS002 Factor Decomposition

AILAF010

Diversified Commodities Seasonal Strategy

  • The strategy demonstrates a residual alpha of 84.09% which is not explained by the 14 factors
  • Key statistically significant factors: Backwardation, Value, Precious Metals Beta, Softs Beta, and Oil Beta

AILAF012

Diversified Commodities Event Driven Strategy

  • The strategy demonstrates a residual alpha of 50.03% which is not explained by the 14 factors
  • Key statistically significant factors: Curve Carry, Trend, Vol Carry, Commodities Beta, EU Power Beta, and Oil Beta
AILAF010 Factor Decomposition
AILAF012 Factor Decomposition

AILAS043

Agriculture Trade Flow Strategy

  • The strategy demonstrates a residual alpha of 91.02% which is not explained by the 14 factors
  • Key statistically significant factors: Backwardation, Vol Carry, Commodities Beta, Softs Beta, and Grains/Oilseeds Beta

AILAS066

Base Metals Seasonal Strategy

  • The strategy demonstrates a residual alpha of 88.18% which is not explained by the 14 factors
  • Key statistically significant factors: Trend, Vol Carry, and Base Metals Beta
AILAS043 Factor Decomposition
AILAS066 Factor Decomposition

Conclusions

  • There is very little correlation between these strategies and standard factors.
  • Base metal and Agricultural strategies do show a slightly positive correlation to the respective sectoral factors.
  • F012 Strategy as designed has negative correlation to most factors as it's expected to perform when there is a sudden market correction.
  • The highlighted factors show which are statistically significant to the underlying strategy.
  • None of the strategies are correlated to the 14 factors chosen, demonstrating truly independent alpha generation.
Correlation Matrix 1
Correlation Matrix 2