130 likes | 216 Views
IDEA Presentation 11/7/09. Panel Discussion. 1. Agenda. Overview of our contribution Omega function Efficiency scores Empirical analysis Conclusion. 1. Overview of contribution:. Most DEA work based on known physical inputs or outputs.
E N D
IDEA Presentation 11/7/09 Panel Discussion
1. Agenda • Overview of our contribution • Omega function • Efficiency scores • Empirical analysis • Conclusion Presentation to Inalytics Australian client update 8
1. Overview of contribution: • Most DEA work based on known physical inputs or outputs. • New theoretical framework needed for portfolio management (allow for non-stationary data; multi-task DMU) • Scope of analysis extended: • Analyse within DMU efficiency across bands • Data needs to go beyond standard Alpha Presentation to Inalytics Australian client update 8
1. Portfolio return analysis – our DMU framework 4 – small position (20%) Own (30%) 5 – big position (10% overweight) Index (e.g. ASX) 2-3 Small CAP underweight Not own (70%) 1 – big caps underweight Presentation to Inalytics Australian client update 8
2. Omega function • Standard metrics (CAPM, etc) assume standard deviation of returns • requires normal distribution: mean and variance are independent random variables • Data exhibits autocorrelation of returns; kurtosis and skewness Presentation to Inalytics Australian client update 8
2.1 Omega function • Omega function corresponds to the sum of the amounts considered as wins, relative to a threshold multiplied by their corresponding probabilities, divided by the sum of the amounts considred as losses, multiplied by their corresponding probabilities • The larger the Omega value, the higher the quality of the bet • Omega functions correspond to relative rate of change of quality of a bet Presentation to Inalytics Australian client update 8
3. Overview: efficiency score • Efficiency is measured by estimating best practice efficient frontiers based of portfolio managers (decision making units or DMUs). • The DMUs on the frontier are considered to be the best practice DMUS because their performance is at least as good as that of other DMUS with similar characteristics. • The efficiencies of other DMUs in the industry are measured in comparison to the efficient frontier. Presentation to Inalytics Australian client update 8
3. Overview – efficiency score • The model comprises multiple inputs and outputs (determined by the user) • The dominating set of ‘efficient DMUs’ consists of at least one DMU observation, and the number of DMU observations for each DMU agent averages about 10. • If a dominating set exists, then the specific DMU observation’s efficiency is measured relative to this set. • If no dominating set exists, the DMU observation is considered self efficient and given an efficiency score of 1. • All efficiency scores range from 0 to 1. Presentation to Inalytics Australian client update 8
3.2 Efficiency model (1) • Inputs • Holding (C) • Max Load difference • Load Diff Sum • Outputs • Construct • Total • Hit rate • OMEGA Presentation to Inalytics Australian client update 8
4.Empirical evidence • Skewness and kurtosis • Omega and alpha scores • Efficiency scores: • Descriptive statistics • Results by DMU • Results Across bands Presentation to Inalytics Australian client update 8
4.1 AUD data - Alpha Skewness(Kurtosis) by DMU Presentation to Inalytics Australian client update 8
4.1 Omega (Alpha) Scores by DMU Presentation to Inalytics Australian client update 8
5. Conclusion • DMU 4 (DMU 5) has lowest (highest) skewness (kurtosis) for Midfield 2 (Attack) • OMEGA varies considerably by DMU; highest average is for midfield and defence • DMU 3 (DMU 1) has highest (lowest) average efficiency (across all bands) • Band 4 is most efficient band (except for DMU1 where it is the least efficient band); Band 2 is least efficient band for DMU1, 2 • Further efficiency analysis can be done assuming variable returns to scale, change over time Presentation to Inalytics Australian client update 8