There are many ways to develop strategies based on the features generated by DLPAL LS.
DLPAL LS does not develop strategies but a set of features that can be used in developing strategies.
The main features are: P-long, P-short, P-delta and S. Values for these features are calculated and displayed on results after running a workspace. These features can be updated automatically. In addition, the program generates train and score files that can be used with these features in machine learning applications.
The program also calculates another set of features based on a group of securities: avgPL, avgPS, Pratio, avgPLS, avgPSS. These features cannot be updated automatically.
There are two major classes of strategies that can be used with the features: fixed and machine learning.
Fixed strategies use the features to trigger entry and exit signals. The simplest possible strategy has the following form:
Buy (cover) if F > a
Short (sell) if F > b
where F is a feature, or function of features, and a, b are some constants.
The simplest strategy involves using P-delta:
Buy (cover) if P-delta > a
Short (sell) if P-delta < -b
In the simplest possible strategy, a=b=0.
A popular strategy involves taking the product of P-delta and S, as follows:
Buy (cover) if P-delta × S > a
Short (sell) if P-delta × S < -b
where a, b are positive numbers usually are set to zero. Ranking may be based on P-delta × S (for example when selecting a few top and bottom stocks from a large universe of stocks.
Alternatively, the strategy may be based on Rank = F, where F is a feature. In this case too, a popular rank is P-delta × S.
The difference the above two strategies is that rank strategies rebalance at every period (every day in daily, every week in weekly timeframes ) whereas the first strategy rebalances positions when the signals are triggered. The difference in turnover can be substantial.
A minimum holding period can be introduced in the above strategies. Usually in the daily timeframe, a minimum holding period of two days will reduce number of trades and commission cost.
There are several other possible strategies and combinations with other indicators. One popular strategy uses the first strategy above with a, b set to 0 with trades entered at the open and exited at the close of the day. This eliminate overnight risk but increases transaction cost.
Another yet possible strategy may involve avoiding long securities that open with a large up gap and short securities that open with a large down gap to avoid price action exhaustion.
Machine learning strategies use the train files to develop models, fixed or abstract, and then the score files to rank new instances and decide which securities to buy and which to short. There are many possibilities and applications in this area but there is also a debate as to whether these models can outperform the simple fixed strategies offered as examples above.
Use in discretionary trading
The signals may be used in discretionary trading in conjunction with additional tehcnical analysis. Long-only, short-only or long/short signals may be considered under discretion from the results of the program.
Examples
Click here for articles with examples.