Trading day - does it matter? Monthly rotational system.

[julo]

Hello,
I have implemented Clenow's rotational system from the Stock on the move book - monthly rotation, calculations on daily bars. The system has five parameters and after optimization phase I reached CAR to MaxSysDD ratio almost 1. Trained on S&P500, past 13 years.
Trading was done at opening of the first trading day of each month, which is usual setup. The problem is that changing trading day has significant impact on strategy performance. Car/MDD will drop to 0.5 - 0.7 even after re-run of optimization phase. This is the reason why I am afraid of trading such system. Intuitively, it looks unstable to me.
Has anyone encountered similar situation? Does it really matter which day one trades? Or is my system likely to be over-fitted?
Thanks for any advice!

 

[qldfrog]

Hello,
I have implemented Clenow's rotational system from the Stock on the move book - monthly rotation, calculations on daily bars. The system has five parameters and after optimization phase I reached CAR to MaxSysDD ratio almost 1. Trained on S&P500, past 13 years.
Trading was done at opening of the first trading day of each month, which is usual setup. The problem is that changing trading day has significant impact on strategy performance. Car/MDD will drop to 0.5 - 0.7 even after re-run of optimization phase. This is the reason why I am afraid of trading such system. Intuitively, it looks unstable to me.
Has anyone encountered similar situation? Does it really matter which day one trades? Or is my system likely to be over-fitted?
Thanks for any advice!

If you follow the market,with the current instability and crash fear,the week end is seen as a risk
what can happen during these 2 non trading days?
So you will notice a yoyo action on monday,buys up and during the week then selling back to cash from Thursday,and repeat..
There is a noticeable day related action but it was not always here.
I believe if backtesting past periods, it will be less marked.
On a monthly system, i would not expect this to make a huge difference
One way to check if your system is too customised is to look at both MC and the actual number of trades.
If you have 1000s of trades,i doubt you could overfit easily, on the other end, and this is an issue with monthly systems, with a few dozen..easily done.
I would be suspicious too.
Fwiw,i tried to change my weekly system day of trade and did not backtest much difference

 

[julo]

If you follow the market,with the current instability and crash fear,the week end is seen as a risk
what can happen during these 2 non trading days?
So you will notice a yoyo action on monday,buys up and during the week then selling back to cash from Thursday,and repeat..
There is a noticeable day related action but it was not always here.
I believe if backtesting past periods, it will be less marked.
On a monthly system, i would not expect this to make a huge difference
One way to check if your system is too customised is to look at both MC and the actual number of trades.
If you have 1000s of trades,i doubt you could overfit easily, on the other end, and this is an issue with monthly systems, with a few dozen..easily done.
I would be suspicious too.
Fwiw,i tried to change my weekly system day of trade and did not backtest much difference

I have around 600 trades during those 13 years. I have checked MC results thoroughly again and found one (possible?) issue. My CAR and MDD are around 20% during backtest on optimized parameters. However, MC result shows Prob(MDD > 20 %) = 90% and Prob(MDD > 30 %) = 50 %. In other words, mean value of MDD is roughly around 30 % which means significantly worse CAR/MDD in average. Am I right?

 

[qldfrog]

I have around 600 trades during those 13 years. I have checked MC results thoroughly again and found one (possible?) issue. My CAR and MDD are around 20% during backtest on optimized parameters. However, MC result shows Prob(MDD > 20 %) = 90% and Prob(MDD > 30 %) = 50 %. In other words, mean value of MDD is roughly around 30 % which means significantly worse CAR/MDD in average. Am I right?

just a quick thought: do you remember the world 13y ago?
now fast track now and do you really think the market are behaving in any way similar to 13y ago? I think that sample is too wide...just my opinion ..I would not get any value from anything pre 2010 and probably pre 2015
but just my opinion

 

[investtrader]

That is a big red flag. Not sure how it could be that different.
Have you tried rebalancing weekly? Intra month drawdowns can be large and I have found that momentum rotational strategies suffer large drawdowns as you are buying the strongest stocks that are great in strong markets, but deadly in risk off markets.

 

[tech/a]

Why are you not buying on triggers on monthly charts on the first day of the month
The same with selling?
it’s a monthly system ——- right?

 

[julo]

just a quick thought: do you remember the world 13y ago?
now fast track now and do you really think the market are behaving in any way similar to 13y ago? I think that sample is too wide...just my opinion ..I would not get any value from anything pre 2010 and probably pre 2015
but just my opinion

It's considerable, but this shouldn't cause the described problem.

That is a big red flag. Not sure how it could be that different.
Have you tried rebalancing weekly? Intra month drawdowns can be large and I have found that momentum rotational strategies suffer large drawdowns as you are buying the strongest stocks that are great in strong markets, but deadly in risk off markets.

I tried rebalancing, but I think that problem is caused by the situation I described in the post #3.

Why are you not buying on triggers on monthly charts on the first day of the month
The same with selling?
it’s a monthly system ——- right?

I do, but I wanted to test the robustness of my system. I won't start trading until I do not fully believe trading system.

 

[Wyatt]

The problem is that changing trading day has significant impact on strategy performance

A few years ago I played around with days of the month using a variant of SOTM and found that the first/last few days of the month produce the best results. I recall doing a lot of googling on the subject and found that is a consistent trait of monthly trend following systems.

You would need nerves of steel to run a monthly system, especially SOTM due to the larger DD. While a weekly timeframe increases the number of trades, it is a lot easier to cope with, but of course, still not easy. I also found little value in the rebalancing feature of the code.

These type of momo strategies with a longer lookback periods, have done fairly poorly on ASX during 2020, although holding their own, so far in 21.

 

[Roller_1]

Hello,
I have implemented Clenow's rotational system from the Stock on the move book - monthly rotation, calculations on daily bars. The system has five parameters and after optimization phase I reached CAR to MaxSysDD ratio almost 1. Trained on S&P500, past 13 years.
Trading was done at opening of the first trading day of each month, which is usual setup. The problem is that changing trading day has significant impact on strategy performance. Car/MDD will drop to 0.5 - 0.7 even after re-run of optimization phase. This is the reason why I am afraid of trading such system. Intuitively, it looks unstable to me.
Has anyone encountered similar situation? Does it really matter which day one trades? Or is my system likely to be over-fitted?
Thanks for any advice!

There is nothing wrong with using daily bars for a monthly system, more data imo.

The first thing i'd suggest is go to Nick Radges twitter/ website and read everything he's written on rotation systems, also sign up to his free newsletter. He is always dropping good articles in this field.

It's called signal luck, you can get it with any system but the longer the time frame the bigger the issue can be. If you have a stock filter or index filter built in this is probably where the 'luck' is coming from.

eg rewind 12 months and the sp500 drops on about the 5th of march, if your lucky the system might exit all positions on the first trading day of March. Using a 200 day SMA you would've done this. It's just a lucky out come in this instance, using a 180 or 220 day you might be holding for the whole month.

As i said he explains it much better than I

 

[julo]

There is nothing wrong with using daily bars for a monthly system, more data imo.

The first thing i'd suggest is go to Nick Radges twitter/ website and read everything he's written on rotation systems, also sign up to his free newsletter. He is always dropping good articles in this field.

It's called signal luck, you can get it with any system but the longer the time frame the bigger the issue can be. If you have a stock filter or index filter built in this is probably where the 'luck' is coming from.

eg rewind 12 months and the sp500 drops on about the 5th of march, if your lucky the system might exit all positions on the first trading day of March. Using a 200 day SMA you would've done this. It's just a lucky out come in this instance, using a 180 or 220 day you might be holding for the whole month.

As i said he explains it much better than I

View attachment 120271

Thanks, this is very relevant information.

 

[Roller_1]

Thanks, this is very relevant information.

You could run a optimization on your parameters such as index or stock filter to see how stable they are.

ie. for a 200 day SMA stock filter, run a optimization from 150 - 250 days, if all of these are profitable and in the same ball park then it should give you more confidence but if say 200 days has a 20% CAR vs 20% DD then 190 days is 12% CAR vs 40% DD then you might have a issue