Risk-Adjusted Performance Attribution and Portfolio Optimisation under Tracking-Error Constraints for SIAS Canadian Equity Fund

This thesis is inspired by the article “Risk-adjusted performance attribution and portfolio optimizations under tracking error-constraints” by Bertrand (2008) together with some hand-on experience gained though managing a portfolio worth over $10 million CAD of the Simon Fraser University endowment Fund for one year. This paper explores the theories of attributing portfolio risk, in the form of tracking-error volatility into asset allocation attributes and stock selection effects in accordance with the arithmetic performance attribution method. Then it applies the same attribution method in calculating the risk adjusted return (information ratio) for a normal portfolio and compare this to a TEV optimal portfolio. We apply the information ratio and tracking-error variance model to the SIAS Canadian Equity portfolio with approximately $4 million CAD in value to test the following: If the SIAS Canadian Equity portfolio sector weights remain the same, what is the expected information ratio? And will this be improved by optimizing the sector weights according to the tracking-error variance frontier? We will then test the robustness of our findings by changing the time period and perform a sensitivity analysis on the estimated expected returns. We will also compare the results with those derived from the mean-variance optimization, by applying mean-variance optimal weights and recalculate the expected information ratio. The findings are as follows: The TEV optimized weights does improve the expected information ratio for a portfolio. This finding is further verified since it gives the same result with different time periods. The sensitivity analysis gives us an interval that the optimized sector weights will be within that interval with 95% probability. Moreover, the comparison to the mean-variance optimized portfolio shows that the tracking-error variance optimization gives less extreme results and is easier to implement, while maintaining a positive expected excess return compared to the benchmark.