Academia.eduAcademia.edu

Figure 9 – uploaded by Peter Stone

See full PDFdownloadDownload figure
Figure 7: Reverse Strategy in PLAT Live-data Competition Round 3, Mar. 18, 2003

Figure 7 Reverse Strategy in PLAT Live-data Competition Round 3, Mar. 18, 2003

Related Figures (17)

Figure 2: March 21st 2003, A “Good” Day profits are the average differences between prices of the order pairs times the number of the simulation ticks. The ideal situation for our strategy is when the end price of a day is very close to the start price. That means we can have our share position close to neutral, and more pairs of our orders executed. On the contrary, if the stock price moves drasti- cally in one direction (either high or low) and doesn’t come back before the end of the day, our strategy will lose money. This becomes clear after some microanalysis. For the sim- ple example we discussed above, when the price drops below $19.00, our buy order will trade and we will have a long po- sition of 100 shares. If the price keeps dropping and never comes back, the value of the 100 shares will be less than $1,900.00 and we will be losing money. Intuitively, such sit- uation will propagate in our strategy, when the price keeps dropping and we keep placing orders. Figure 2 and Figure 3 are the price charts of MSFT on March 21st and March 24th, 2003, which are examples of a “good” day and a “bad” day towards the strategy respectively.
Figure 2: March 21st 2003, A “Good” Day profits are the average differences between prices of the order pairs times the number of the simulation ticks. The ideal situation for our strategy is when the end price of a day is very close to the start price. That means we can have our share position close to neutral, and more pairs of our orders executed. On the contrary, if the stock price moves drasti- cally in one direction (either high or low) and doesn’t come back before the end of the day, our strategy will lose money. This becomes clear after some microanalysis. For the sim- ple example we discussed above, when the price drops below $19.00, our buy order will trade and we will have a long po- sition of 100 shares. If the price keeps dropping and never comes back, the value of the 100 shares will be less than $1,900.00 and we will be losing money. Intuitively, such sit- uation will propagate in our strategy, when the price keeps dropping and we keep placing orders. Figure 2 and Figure 3 are the price charts of MSFT on March 21st and March 24th, 2003, which are examples of a “good” day and a “bad” day towards the strategy respectively.
Figure 3: March 24th 2003, A “Bad” Day Based on the market making strategy we are using, our agent tends to get into a large short position if the price goes up consistently; on the other hand, a large long position if the price goes down stiffly. Either of these situation will lead to a loss in value, according to the discussion above. Further- more, the value will be at the mercy of the huge position, and the profit accumulated by our strategy will be domi- nated by it. That is, for instance, if we hold a lot of volume and the stock price goes down, our resulting losses dominate the small profits we get from each matched pair of trades. To mitigate such a problem, we can try to prevent our holdings from going to one extreme or the other. We have designed two methods, which are based on the two most important characters of an order - price and volume, to achieve a bet- ter control on our share position.
Figure 3: March 24th 2003, A “Bad” Day Based on the market making strategy we are using, our agent tends to get into a large short position if the price goes up consistently; on the other hand, a large long position if the price goes down stiffly. Either of these situation will lead to a loss in value, according to the discussion above. Further- more, the value will be at the mercy of the huge position, and the profit accumulated by our strategy will be domi- nated by it. That is, for instance, if we hold a lot of volume and the stock price goes down, our resulting losses dominate the small profits we get from each matched pair of trades. To mitigate such a problem, we can try to prevent our holdings from going to one extreme or the other. We have designed two methods, which are based on the two most important characters of an order - price and volume, to achieve a bet- ter control on our share position.
After experimenting with a number of fixed gaps, it became apparent that our method is sensitive to the magnitude of the gap. Furthermore, due to the difference in prices and trading patterns for different stocks, it is likely that this value would need to be tuned anew for each different stock. Here we introduce a method for choosing the gaps dynam- ically based on the current order books. We expect that placing orders with reference to the existent orders in both the buy and sell books will be more robust to a variety of trading scenarios. If and when PLAT expands to include more than one stock we plan to test this hypothesis explic- itly. Our strategy currently takes in a parameter n, which varies from 1 to the number of the existent orders. The or- ders will be placed immediately in front of the nth order in both queues. For example, the buy order we insert will be at $0.0001 higher than the price of the nth buy order, while the sell order will be at $0.0001 lower than that of the nth sell order. Table 1 shows a snapshot of the first five orders in both buy and sell order books where the order book ori- ented market making strategy is used with n=2. The orders in bold outline are the virtual orders placed in the simulator by our strategy, while the others are the real orders in the market. Table 1: Buy and Sell Order Books
After experimenting with a number of fixed gaps, it became apparent that our method is sensitive to the magnitude of the gap. Furthermore, due to the difference in prices and trading patterns for different stocks, it is likely that this value would need to be tuned anew for each different stock. Here we introduce a method for choosing the gaps dynam- ically based on the current order books. We expect that placing orders with reference to the existent orders in both the buy and sell books will be more robust to a variety of trading scenarios. If and when PLAT expands to include more than one stock we plan to test this hypothesis explic- itly. Our strategy currently takes in a parameter n, which varies from 1 to the number of the existent orders. The or- ders will be placed immediately in front of the nth order in both queues. For example, the buy order we insert will be at $0.0001 higher than the price of the nth buy order, while the sell order will be at $0.0001 lower than that of the nth sell order. Table 1 shows a snapshot of the first five orders in both buy and sell order books where the order book ori- ented market making strategy is used with n=2. The orders in bold outline are the virtual orders placed in the simulator by our strategy, while the others are the real orders in the market. Table 1: Buy and Sell Order Books
Figure 4: MSFT Stock Price: Dec. 20, 2002 The reason that the reverse strategy makes profits in many kinds of market is that stock market prices are not constant and in fact undergo frequent changes in direction, rather than moving consistently in one direction. On most trading days, there are a lot of small spikes in the stock price chart. Figure 4 shows such an example. Under these conditions, we expect to make profits with the reverse strategy, as de- scribed in the next subsection.
Figure 4: MSFT Stock Price: Dec. 20, 2002 The reason that the reverse strategy makes profits in many kinds of market is that stock market prices are not constant and in fact undergo frequent changes in direction, rather than moving consistently in one direction. On most trading days, there are a lot of small spikes in the stock price chart. Figure 4 shows such an example. Under these conditions, we expect to make profits with the reverse strategy, as de- scribed in the next subsection.
Table 9: The Calculation of the Reverse Strategy: Linear Model
Table 9: The Calculation of the Reverse Strategy: Linear Model
Table 8: The Reverse Strategy over One Cycle
Table 8: The Reverse Strategy over One Cycle
Table 7: The Basic Strategy over One Cycle. the price of MSFT oscillates consistently in a wave pattern around a constant price p. We call the microvariation of the price (the wave’s amplitude) during the trading day 6. Fig- ure 5 part (a) illustrates such a price trend. We also assume that all orders placed in buy queues and sell queues will be matched at some point during the trading day (perhaps im- mediately). In our analysis, we assume that the start price is p—6. The start price is not crucial to the analysis since if it’s not p — 6, we can ignore the first several ticks until we have the price p — 6. Table 7 and Table 8 show the cal- culations of the changes in holdings, cash and value for the basic and reverse strategy in one cycle. Each time we place a sell order or buy order, the volume size is v.
Table 7: The Basic Strategy over One Cycle. the price of MSFT oscillates consistently in a wave pattern around a constant price p. We call the microvariation of the price (the wave’s amplitude) during the trading day 6. Fig- ure 5 part (a) illustrates such a price trend. We also assume that all orders placed in buy queues and sell queues will be matched at some point during the trading day (perhaps im- mediately). In our analysis, we assume that the start price is p—6. The start price is not crucial to the analysis since if it’s not p — 6, we can ignore the first several ticks until we have the price p — 6. Table 7 and Table 8 show the cal- culations of the changes in holdings, cash and value for the basic and reverse strategy in one cycle. Each time we place a sell order or buy order, the volume size is v.
Figure 5: Different price trend models: (a) shows the initial model in which the price oscillates consistently in a wave pattern around a constant price. (b) extends the model to allow for the price to oscillate around a line with non-zero slope.
Figure 5: Different price trend models: (a) shows the initial model in which the price oscillates consistently in a wave pattern around a constant price. (b) extends the model to allow for the price to oscillate around a line with non-zero slope.
Table 10: PLAT Live-data Competition Round 1, Blue Pool. (Team 5: our reverse strategy) In the first round of the PLAT live competition, the market- making strategy was placed in the red pool, and was Team 8. The reverse strategy was placed in the blue pool and was Team 5. Table 10 and Table 11 show the results from the first round. Team 11 in the red pool was disqualified after three trading days because of repeated and extreme viola- tion of the share position limit. Team 7 had third-place earnings in the red pool, but was disqualified for its share position limit violations three times in five days. The red pool qualifiers were Teams 12, 8 (our market-making strat- egy) and 13. The blue pool qualifiers were Teams 4, 5 (our reverse strategy) and 1.
Table 10: PLAT Live-data Competition Round 1, Blue Pool. (Team 5: our reverse strategy) In the first round of the PLAT live competition, the market- making strategy was placed in the red pool, and was Team 8. The reverse strategy was placed in the blue pool and was Team 5. Table 10 and Table 11 show the results from the first round. Team 11 in the red pool was disqualified after three trading days because of repeated and extreme viola- tion of the share position limit. Team 7 had third-place earnings in the red pool, but was disqualified for its share position limit violations three times in five days. The red pool qualifiers were Teams 12, 8 (our market-making strat- egy) and 13. The blue pool qualifiers were Teams 4, 5 (our reverse strategy) and 1.
Table 12: PLAT Live-data Competition Round 2. (Team 5: our reverse strategy; Team 8: our market-making strategy)
Table 12: PLAT Live-data Competition Round 2. (Team 5: our reverse strategy; Team 8: our market-making strategy)
Table 11: PLAT Live-data Competition Round 1, Red Pool. (Team 8: our market-making strategy)
Table 11: PLAT Live-data Competition Round 1, Red Pool. (Team 8: our market-making strategy)
Figure 6: Reverse Strategy in PLAT Live-data Competition Round 3, Mar. 17, 2003
Figure 6: Reverse Strategy in PLAT Live-data Competition Round 3, Mar. 17, 2003
Table 14: Historical Experiments Besides the live-data competition, we also did some histor- ical experiments with our agent strategies. Table 14 shows the results. Column 2 shows the value of the market-making strategy and Column 3 shows the value of the reverse strat- egy. Both strategies ran on the historical dates separately. From Table 14, we can see that both strategies lost money some days and made profits on other days. We did our ex- periments on dates from Mar 21, 2003 to Apr 03, 2003. We choose these dates before seeing the results and report all of the results here. The MSFT stock price endured high vari- ability during these days due to the war affairs. We hope the performances of our agent strategies will be much better if change of the MSFT stock price is stable and consistent. Meanwhile, Risk control mechanisms should be emphasized to make the strategies more adaptable and profitable.
Table 14: Historical Experiments Besides the live-data competition, we also did some histor- ical experiments with our agent strategies. Table 14 shows the results. Column 2 shows the value of the market-making strategy and Column 3 shows the value of the reverse strat- egy. Both strategies ran on the historical dates separately. From Table 14, we can see that both strategies lost money some days and made profits on other days. We did our ex- periments on dates from Mar 21, 2003 to Apr 03, 2003. We choose these dates before seeing the results and report all of the results here. The MSFT stock price endured high vari- ability during these days due to the war affairs. We hope the performances of our agent strategies will be much better if change of the MSFT stock price is stable and consistent. Meanwhile, Risk control mechanisms should be emphasized to make the strategies more adaptable and profitable.
Figure 8: Market Making Strategy in PLAT Live-data Com- petition Round 2, Mar. 06, 2003
Figure 8: Market Making Strategy in PLAT Live-data Com- petition Round 2, Mar. 06, 2003
Table 15: Experiments on Variations of Market Making Strategy
Table 15: Experiments on Variations of Market Making Strategy
Figure 9: Market Making Strategy in PLAT Live-data Com- petition Round 2, Mar. 07, 2003
Figure 9: Market Making Strategy in PLAT Live-data Com- petition Round 2, Mar. 07, 2003
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved