Statistical Arbitrage Team Leo, Ying, Yandong, Xing
Agenda Debugging Sorry to disappoint, but that’s all we did…
Data Prices regenerated using CRSP’s adjusted returns – Match Yahoo Finance prices for 5-6 randomly picked stocks – Several small consistency checks » Input from all group members Want to match the 2 ways of computing prices – Almost there…
PCA Issues Preliminary experiment – Get minimum/maximum/mean volatility – Get sum of weights in the portfolios Debug the PCA code – What are the value ranges? – Include volatility in the calculation? – Portfolio weights should always add to one
PCA Eigenvector Weight Sums
Volatility Volatility Histogram
Fig. 3 Experiment Comparison Paper’s Current Previous
With and Without dividing by Volatility With Div. Volatility Without Div. Volatility
PNL – Paper’s x Our Results Paper’s Ours
PNL – Current x Previous Results Previous Current
PNL – Paper’s x Our Results Paper’s Current Previous
Trading – Debugging Plan Debug! Add transaction costs Make sure it is market neutral Compute Sharpe ratio Display positions Debug with “deterministic” dummy ETF/stock Signals – Are they generated correctly? Merge experiments in single code base
ETFs – Action Plan We’ve started matching Stock with ETF – Using GICS as proxy Built dictionary Only part of the Stocks have GICS code – About ~1/2 of the securities – Use another proxy (NAICS/SIC)? Change Simulations to incorporate ETFs – Straightforward once the rest is working
Suggestions?