150 OLPS: A TOOLBOX FOR ONLINE PORTFOLIO SELECTION
Figure A.3 Various components of the Experimenter.
When the execution is over, the Results Manager shows all the basic performance
metrics of the algorithms. Since we have two different managers—one for analyzing a
single algorithm and one for comparing multiple algorithms—we made two different
Results Managers.
Results Manager 1 The first Results Manager for the Algorithm Analyzer is shown
in Figure A.4. The table in the window quantifies the results of the algorithm as
compared to the basic benchmarks. The numbers from this table can directly be
copied and pasted. There is a large graph space that displays the information on a
particular attribute selected in the left column.
Returns It contains information about the daily performance of the algorithm.
The user can choose to view the cumulative returns and the daily returns. The option
of a log (base 10) plot is provided for easier visualization when the difference in
performance of the algorithm and the benchmarks is significantly high.
Risk Analysis There are five metrics to evaluate the risk and risk-adjusted returns
of the algorithm. They are the Sharpe Ratio, Calmar Ratio, Sortino Ratio, Value at
Risk, and Maximum Draw down.An input box called Window is provided next to each
metric. The purpose of the window is to analyze the consistency of the algorithm,
instead of just the final result. For example, entering 252 in the Sharpe Ratio Window
will plot a graph of the Sharpe Ratio of the algorithm for time period t −252 to t, for
all t. When the window size is large such that t is less than the window size, then the
computation starts from t = 1. The risk metrics are assumed to be zero for the first
50 time periods. This has been done to avoid extreme values due to lack of data in
the initial periods.
Portfolio Analysis The Portfolio Allocation shows the distribution of wealth
allocated to each asset by the algorithm. The Step by Step helps us look at the port-
folio allocation for any particular given day. Lastly, we have a portfolio Animation
that accepts an input called Window. Visualizing portfolio changes based on daily
frequency can be overwhelming and difficult to interpret, especially when the daily
T&F Cat #K23731 — K23731_A001 — page 150 — 9/28/2015 — 20:46