PENN AUTOMATED TRADING PROJECT

HISTORICAL DATA COMPETITION RESULTS, DECEMBER 2002

SIMULATION DAYS: DECEMBER 7-12, 2002

HISTORICAL TRADING DAYS: NOVEMBER 13, 19, 25

STOCK TRADED ON PXS: MSFT


BRIEF DESCRIPTION: Each team was given 3 of the 6 days during which they were allowed to run only one simulation at any time. Since each simulation takes between 8 and 12 hours of real time, there was limited opportunity for teams to get multiple runs on each day and report their best results. Each team was required to remain inside a 100,000 share position limit at all times, and teams were required to trade the strategies they had previously developed and described in advance of the historical dates being chosen and announced, and the data released.

BRIEF ANALYSIS: Perhaps the most interesting outcome was the overall strength of strategies that in some way used the order books in the computation of when to place orders, and/or in what volume. The top four strategies all shared this property, though their uses of the order book data was varied. There were a number of variants of the SOBI (Static Order Book Imbalance) strategy we studied earlier in the term. Some of these variants used a SOBI-like strategy as their main approach, while others used quartile averages as a modifying factor on top of some other strategy (such as market-making).

A number of more traditional technical trading strategies, such as those computing breakouts from moving averages, were also profitable overall. Perhaps the most common source of underperformance was the failure to saturate the allowed risk, which is discussed in the News and Notes.

KEY AND COMMENTS:

RANK TEAM MEMBERS DESCRIPTION NOV 13 VALUE (MAX SHARES ) NOV 19 VALUE (MAX SHARES) NOV 25 VALUE (MAX SHARES) 3-DAY VALUE COMMENTS
          1 Mahesh SOBI variant: uses standard deviation of buy and sell queues 15.1 (70.0) 66.1 (80.0) 44.3 (45.0) 125.5 A simple strategy with a good story; decent risk saturation
          2 Lowenstein, Rowin, Schwartz, Skolnick Market-maker with volume moderated by SOBI signal 22.0 (80.0) 30.0 (105.0) 70.0 (99.0) 122.0 More SOBI than MM?; good (over)saturation of risk
          3 Lao, Leung, Wong Case-based learning on top of basic SOBI 42.7 (47.1) 16.2 (16.1) 11.8 (30.9) 70.7 Closest to original SOBI; could have won with better risk saturation?
          4 Yu, Zhang Short-term trend tracking, long-term breakouts moderated by SOBI 47.3 (72.5) 21.1 (41.0) -19.9 (112.0) 48.5 Interesting mixture of approaches; did queue info help here?
          5 Morgovsky, Sharma Moving average trends 39.7 (60.0) 13.3 (80.0) -7.7 (60.0) 45.3 Highest-ranking strategy not looking at queues
          6 Jawaid, Shah Conjunction of range breakout and moving average crossovers 40.8 (110.9) -6.2 (107.1) -0.8 (100.2) 33.8 Combo of two common tech trading strategies
          7 Kadakia, Vora Market-maker adjusted by volatility 12.5 (57.4) 3.6 (40.0) 9.0 (82.0) 25.1 Might be only "true" market maker in group
          8 Yu (U. Texas) Simple trend strategy -2.7 (90.0) 6.9 (90.0) 20.7 (40.0) 24.8 Not bad for such a simple idea
          9 Angel, Hausman Intersecting geometric trend lines 3.8 (13.0) 0.4 (5.8) 3.1 (12.3) 7.3 Positive earnings each day, but greatly undersaturated risk
         10 Chi, Koltunov, Luss, Trinh SOBI moderated by volatility -5.8 (55.0) -0.1 (1.4) -0.1 (0.9) -6.0 Poor saturation on last two days
         11 Chavez Crossover moving average with added features -5.9 (3.5) -1.7 (3.0) -1.2 (2.5) -8.8 Plots show extreme volatility in position; very low saturation