3 attempts to enhance spy611 effectiveness

[Dan Bikle]

People,

I tried out 3 ideas tonight to enhance the effectiveness of spy611.

Idea 1:

Add a feature called recent_volatility which is simply the standard deviation of pct_gain over the past 40 days.

Result:

Failure.  Adding this feature reduced effectiveness.

Idea 2:

Add a feature called horizontal_distance which is the number of days since the last 52 week high.

Result:

Failure.  Adding this feature reduced effectiveness.

Idea 3:

Add weight to in-sample data which is closer in time to the out-of-sample data.

I implemented this by duplicating rows in the in-sample data if the rows were "younger".

Logic Details:

Rows aged between 0 and  5 years are duplicated 4 times.
Rows aged between 5 and  10 years are duplicated 3 times.
Rows aged between 10 and  15 years are duplicated 2 times.
Rows aged between 15 and  20 years are duplicated 1 time.
Rows aged between 20 years and 25 years are not duplicated.

Result:

Some success.
Backtesting effectiveness increased by about 20%

Dan