Rotational Trading: how to reduce trades and improve returns

With this post I want to share some research I’ve done in the area of rotational trading (RT). In RT systems stocks or ETFs are ranked according to one or more properties. You ride the best stocks as long as they are among the best stocks, then you change horses and go again. So far so good. Unfortunately some times you change stocks just to see that the stock you sold is doing better again and raising in your ranking. As we know the market has a certain amount of noise that can’t be predicted or modeled, hence stocks will raise or fall just because of that noise. This represent two challenges: trading cost and opportunity cost. Let me present you some ways how to reduce the impact of that volatility in ranking.

.

Background

There is a number of ways to reduce rank volatility.

1.) Trade only once a week / month

2.) Different criteria for enter / exit a trade

3.) Smooth the rank over the last couple of bars

4.) Combination of above

Test Environment

I run all tests on a survivorship free Nasdaq100 index ( January 2001 – June 2011)
All trades are close/close without slippage / commission
10 portfolio positions (10% allocation), returns are reinvested
Data is adjusted for splits and cash dividends
Stocks ranked according to their Rate of change over the last 200 days (ROC200)
Ranking is re-calculated every day

Test 1.) Weekly rebalancing

In this test we compare daily ranking vs. weekly ranking. For weekly ranking I decided to re-balance at the last trading day of a week. Important to notice: also for weekly re-balancing daily bars are used for trading and rank calculation.

Number of trades have been reduced by 50+% while we see a modest improvement in absolute returns.

Test 2.) different entry/exit rank

In this test a trade is entered if the stock is among the best performing stocks and sold if it drops bellow a certain rank (10, 20, 30). Of course new stocks can only be bought if a portfolio slot has been (=an existing position drops bellow 10/20/30).

Interesting enough the absolute returns have improved significantly while reducing trade frequency by about 90%.

Test 3.) Rank smoothing

Instead of using the most recent rank value for taking trading action, the rank gets smoothed with it’s most recent values. For this test I applied to different method, simple moving average (MA) and exponential smoothing average (EMA). Rank calculation and trades are done daily.

Applying a ten day EMA seems to be most promising in this case.

Test 4.) Combination

In a last test we will look at combining different methods.

Summary

The presented methods provide valid ideas to reduce market noise and improve returns for a simple rotational system. Considering that trading cost (slippage + comissions) has not been part of this test, the absolute return difference is even more significant. The applied methods reduced trades by about 50-90%. So which one is the best one? I prefer weekly rotational trading for my momentum system. However, at the end it’s a matter of personal trading style and the specifics / efficiency of your ranking algo.

Please let me know your opinion and ideas in the comment section bellow.

- Frank

About these ads

Comments

oliver says:

July 6, 2011 at 15:24

Hi,
Good post and good thinking.
I have thought about RT trading quiet a bit but i have a problem programming it as i am not yet proficient enough, would you be able to help with the basic programming idea behind ranking?
All the best,
Oliver

- Frank Hassler says:
  
  July 6, 2011 at 15:37
  
  Hi Oliver,
  
  I just sent you an email.
  
  Frank
  
bgpl says:

July 6, 2011 at 15:35

hi Frank,

Doesn’t this depend on what metric you are trading in the first place ?
ROC(200) can be noisy when you compare several stocks since it closely depends on today’s price, which is noisy in the first place i.,e C/Ref(C,-1) – 1;
I think if you use something else which is slow varying and already smoothed such as
TSI() the daily may not lead to so many trades.
So, not sure if your experiments are more a choice of exits conditions, or only reflecting the noisiness of the entry.
Just an observation, not to detract from your post in anyway, which as always, is thought provoking and well presented !

regards,
bgpl

- Frank Hassler says:
  
  July 6, 2011 at 15:38
  
  Hello Badri,
  
  agree: ROC200 is noisy as it’s not considering a stocks inherit volatility. However, I wanted to provide some useful ideas.
  
  Have a great time my friend…
  
  Frank
  
lantama says:

July 6, 2011 at 16:33

Frank, good work as always. Where did you get the underlying data from? I mean dividend and split adjusted end-of-day data?
Thank you
Andreas

- Frank Hassler says:
  
  July 6, 2011 at 16:36
  
  Hi,
  
  Norgate / premiumdata provides cash adjusted data with their new service (look at one of my last posts)
  
  Frank
  
Pete Kasper says:

July 7, 2011 at 23:40

Very interesting Frank. The drawdown, however, takes my breath away.

PK

- Frank Hassler says:
  
  July 8, 2011 at 07:34
  
  Hi Peter,
  
  sure, I wouldn’t want to trade this. It’s not a system. But some ideas can be taken from the post to build a better system.
  
  Frank
  
Sunil says:

July 7, 2011 at 23:58

Frank, How about using a a market filter to exit all positions and move into cash? Of course this will reduce the CAR a little but the Max DD should improve quite a bit.

- Frank Hassler says:
  
  July 8, 2011 at 07:33
  
  Hello Sunil,
  
  sure these things can be done. The aim of this post has not been to create a system. I rather wanted to show how noise can impact the ranking and what to do about.
  
  Frank
  
Mike says:

October 26, 2011 at 06:23

Read this post many times! Ideas are great. Was wondering if you ever come across the concept of dividing ROC by volatility to form a risk adjusted measure of relative strength? I did some test on STD vs ATR. STD was the superior one and shorter term volatility seem to be better too. I also tested it across TAA so it seems robust.