Take Profit Seasonality on Overnight Trading

Take Profit Seasonality on Overnight Trading

I have always thought of a simple exit at the next days open for nightly trades on ES future to be the most simple and non-optimized exit. There is no risks of data-mining biases. This is what I always do until now at Nightly Patterns.

Overnight behavior changes with the market’s regimes of volatility. Nobody can argue that volatility does not strongly differ across the year. Seasonality looks to be one of the strongest filters for a take profit order optimization.

I started with a simple backtest considering all nights back from June the 15th 1998 to December the 4th 2015 (time of writing). I considered all nights of the week except Friday and the last day of each holidays. Many times big gapping nights occur after the markets are closed on weekends or because of holidays. In these situations a simply “hold till the open” approach looks better to deeply exploit the weekend/holiday potential.

On the other hand, many traders doesn’t like the risk of being invested during these long stock market pauses.

Firstly, trading all nights by going LONG only (except Fridays and last day before holidays) from 4.00 PM EST to 9.00 PM EST of the next day, resulted in a 45,85% net gain.

Secondly, I tried to data-mine the best take profit order. Using a fixed take profit of 0,4%, the backtest showed a net gain of 116,65%: more than 2 times the baseline system with no take profits. It seems that by just adding a small take profits it boosts returns higher. I thought of a 1% take profit to work better but it’s not. Backtesting the numbers 0,4% take profit is strongly the best ever.

Moving onto the next step, I added a simple monthly seasonal filter. I optimized the best take profit for each month as shown in this table:

MonthJanFebMarAprMayJunJulAugSepOctNovDecTotal net gain %
Take Profit %0,40,350,650,40,350,50,710,9520,80,55
Net gain %14,849,3415,1913,218,4823,3918,55-1-213,7217,9734,61166,30

It’s clear how the historically mostly difficult months for the markets like August, September, and October need larger take profit orders to face the markets than more quiet winter/spring months.

This is the first big anomaly I found looking at seasonality with take profit orders.

Another unexpected finding is the impact of the “day of the week” filter. I looked at all Mondays, Tuesdays, Wednesdays, and Thursdays. Monday showed a different behaviour whilst the other three days result somewhat equal.

DayTake Profit %130,85

So, I backtested separately all Mondays through different months:

MonthJanFebMarAprMayJunJulAugSepOctNovDecTotal net gain %
Take Profit %1,61,25NO0,650,90,50,452,20,3560,80,55
Net gain%3,954,516,333,871,4910,95,124,963,3212,46,389,2672,49

Results are totally different from the all days of the week table. This apparent randomness could really be random but, just because Mondays look so different they need to be taken apart from the other days. And at the moment, there’s nothing better to use than these data-mined results.

And here’s the other three days of the week (Tuesday, Wednesday, Thursday):

MonthJanFebMarAprMayJunJulAugSepOctNovDecTotal net gain %
Take Profit %0,40,350,351,150,350,450,710,950,40,80,55
Net gain %13,919,7312,213,99,5513,2915,15-0,86-1,3510,7111,5925,90133,72

It’s funny how November and December, the most strongly bullish months of the year, showed the same take profit for all the week.

To sum up, by adopting a different take profit for Mondays, and the other three days together following the last two tables we get a total net gain of 206,21%, more than 4 times the baseline amount of 45,85% with no take profit at all.

It could be just result of data mining bias, but the seasonality is there and the numbers point at it.

About the Author Marco Simioni

  • Riz says:

    Dear Sir,

    It would be great if you could offer this course online for people in different countries

  • >