Use Case 4: Dynamic Trader Profiling (DTP) After all the time spent in fundamental and technical analyses, sooner or later comes the time for trade execution. Dynamic Trader Profiling is an automatic technique designed to help traders understand other traders on the other side of the trade and thereby maximize the profit from trading. The technique has potential both as a handy tool for traders or as an automatic trading program that could potentially replace human traders. The concept of DTP stems from the commonly applied method of comparing technical analyses and fundamental analyses. The smartest investors use fundamentals as a stock screener, and technicals as the trading signal. Fundamentals can give investors a fuzzy idea of whether a company is under or over valued but fundamentals can never tell investors when is the best time, not only just a good time, to trade a stock. Dynamic trader profiling attempts to profile the traders behind the trades. Instead of predicting the best time to buy a stock, or the maximum profit, the goal of DTP is to pin-point local highs and lows. The following details the theory behind DTP:

 

Assumption 1: There are two main types of investors/traders: Retail, and Institutional.

Assumption 2: Retail trading does not significantly impact prices and volumes. This implies that DTP must focus on institutional trading.

Fact 1: Institutional traders are generally highly educated.

Fact 2: An educated person keeps facts in mind for a period of time.

Fact 3: A person can track limited screens during any given time.

Fact 4: An institutional trader has a target price range and size range in mind before trading.

Fact 5: An institutional trader tracks a given stock for a while before executing the trade.

Fact 6: At any time at any price, the trader only has three reactions: Buy, Sell, or Wait

Fact 7: At any price, for some time period, the trader keeps his Buy, Sell, or Wait interest.

Conclusion 1: for a given stock, there are limited institutional traders tracking it, and each of them tracks the stock for a period of time during which they have some interest in the stock. Every move in the quote system will have some impact on their interests.

Conclusion 2: What moves a stock’s price is the timing of a combination of buy interests and sell interests of institutional traders.

Conclusion 3: When the surplus of buy interest over sell interest decreases to zero, we find a local maximal price. Similarly, we can find local minimal price.

Historical observation 1: Stock is more stable and more predictable between 10 AM and 3:30 PM

Historical observation 2: Between 10 AM and 11 AM, stock hits a high or low for the period between 10 AM and 3:30 PM

Problem 1: How to measure the surplus?

DTP technique: List scenarios of all stocks’ real trading sequences. Using data mining techniques, predict in real time the local highs and lows for each stock.

 

In the following we detail a sample scenario employing DTP.

 

On 07/25/2001 it was a slow bear market, and AGR.A, Agere Systems had taken a beating the previous day along with Lucent, its parent company, when Lucent announced poor earnings. One day later on the 25th Agere announced its earnings with lowered expectations for the next quarter. That was very bad news. Agere had already experienced two price declines in previous months. Meanwhile, Agere had recently made its IPO with promising technology, but its IPO price had been reduced several times.

 

The DTP technique initializes Agere with a sell recommendation, with short term buy interests when it is between its IPO price of $6 and its previous close low of $5. These values are preset before the market opens. Once the market has opened, DTP records all trading sequences. For example, on the morning of July 25th, Agere was down sharply. DTP can be applied effectively after a stock undergoing a sharp decline in price has begun to stabilize.

 

The table below exemplifies a simplified trade sequence that will aid in understanding DTP in this scenario involving Agere Systems.

 

Time	Bid Price	Bid Size	Ask Price	Ask Size	Last Trade	Trade Size
10:xx:15	5.2	10000	5.22	20000	5.22	50000
10:xx:20	5.2	9500	5.22	20000	5.2	500
10:xx:30	5.2	10000	5.22	16000	5.2	4000
10:xx:45	5.2	9500	5.22	20000	5.2	500
10:xx:50	5.2	10000	5.22	20000	5.2	0
10:xx:00	5.17	20000	5.19	10000	5.17	10000

 

At 10:xx:20 AM, there was a trade of 500 shares at the bid price, which means it was an active sale, but the size was small compared to trades in the previous 30 minutes. Following this, the bid size became 10000 (up from 9500). This could mean two things: a small buy order was put in at 5.2 or a fresh new bid order of 10000 along with a cancellation of the previous order. What happened at 10:xx:30 confirms that some trader or a combination of traders used fresh new bid and ask orders and canceled previous uncompleted orders. That signified that the price had reached the level that the trader(s) had predetermined was a level at which they wanted to trade (cf. Fact 4 above). The volume also was down considerably at that point in time, indicating a reduction in noise (e.g., decrease in day-traders’ activity).

 

At 10:xx:00, the price dropped again, this time to 5.17, but the bid size was much larger and ask size much smaller. Then the price continued dropping with little volume. When it reached 5.05 several minutes later (not shown in the table), it came back up and stabilized with bid at 5.10 and ask at 5.15 with only about one trade per minute even though it was a busy trading day. That means there was no significant interest in selling in such a low price range. This confirms that the 07/25/2001 local low was 5.05, because most stocks reach their local lows and local highs between 10 AM and 11 AM (cf. Historical observations 1 and 2 above).

 

A data mining program based on DTP could automatically set up a bid order at 5.11, one cent ahead of blocks that tend to be traded on .05 boundaries (e.g., 5.10). After the order closed, the program could automatically look for a local high to dump the stock due to a long term sell rating on Agere.

 

Dynamic Trader Profiling can be used to assist traders and analysts to increase the coverage per trader/analyst. In other words, by automating techniques such as this, traders can cover many more stocks than otherwise. A  DTP data mining program could automatically send trade alerts to traders to execute a trade and also to analysts to research a given stock. The strength of this approach, however, lies less in its novelty than in the fact that the DTP technique described above can be done completely automatically based on data mining techniques. For example, even an order could be placed automatically. A DTP-based data mining program could track the performance of the trading strategy, and, for example, place a larger order if DTP performed well previously under similar conditions.

 

The following examples show actual investment gains and current holdings using a combination of the two techniques described above, DTP and FSC:

 

1. Bought AGR.A at 5.36 on 06/22/2001, and sold it at 6.94 on 06/29/2001 with 30% gain in 1 week.
2. Bought AGR.A at 5.11 on 07/25/2001 and sold at 5.74 on 07/27/2001 with 12% gain in two days
3. Bought CORL at 2.76 on 07/02/2001 and at 2.89 on 07/03/2001, and sold at 3.84 on 08/03/2001 with a 35% gain in one month

 

One disadvantage Dynamic Trader Profiling has is the volume it trades must be less than about 5% of the average daily volume of the stock it trades. DTP won’t work when DTP becomes the market maker. One of the advantages of DTP is that it only needs two to four weeks of real time trade sequences for sampling to enable it to start making predictions. Until now, limited resources have not allowed the sampling of data for every stock. However, if all stocks are sampled and only 10% are predictable by a DTP-based data mining program, it would be sufficient to keep a multibillion-dollar fund quite busy trading.