Questions in the mean reversion course

AI & Machine learning
1

I am very new to quant trading learning and I have a few problems in implementing the strategies reagrading to the mean reversion course. 


  1. In the couse it says to do the CADF test we have to do a regression analysis test using first 90 days data to find the hedge ratio. Suppose I had developed a trading model for it and as time goes, do I need to update the hedge ratio based on any signal or any number of days elapsed? Do I need to discard the old data then?


  2. In the course it says even if the stationarity result is not good the pairs might work well too given that the half life is short. In what benchmark/balance could we determine it? (e.g. Cointegraion with say only 70% level of confidence can work well?) 


  3. Do you recommend running ADF/johansen test to check stationarity daily if the trading strategy is on daily basis?
2

 

1) In the couse it says to do the CADF test we have to do a regression analysis test using first 90 days data to find the hedge ratio. Suppose I had developed a trading model for it and as time goes, do I need to update the hedge ratio based on any signal or any number of days elapsed? Do I need to discard the old data then?



You can run the CADF test on a rolling basis. For example, say at the start of each month and if the pair is cointegrated then trade for that particular month. The frequency of each month is only for illustration purpose. The idea is running CADF test too frequently is not required as cointegration is associated with long-term properties of time series and running CADF test after a long period of time, might lead to missing out on recent price dynamics. You have balance between the two.

2) In the course it says even if the stationarity result is not good the pairs might work well too given that the half life is short. In what benchmark/balance could we determine it? (e.g. Cointegraion with say only 70% level of confidence can work well?) 


You can think of these two as independent filtering condition for identifying pairs. For example, if cointegration > 90% or half-life < 5 days.

3) Do you recommend running ADF/johansen test to check stationarity daily if the trading strategy is on daily basis?


Refer to answer 1 above

I hope this helps.
3

Thanks for your kind reply Ishan!



I feel very frustarted on one thing: if my strategy is built on the logic of mean-reversion making use of the cointegration properties of a pair, so I would think if on some day this relationship breaks (reflected by daily ADF test), I should discard the trading strategy because the logic behind does not hold anymore. However both the course material and in my backtesting suggest the opposite - break in cointegration at certain time period doesn't mean that I would have a loss in daily return in that period.


  1. What is the flaw of my assumption?
  2. How are professional traders monitor their mean-reversion trading strategy to ensure the rationale behind is right? I mean, what signal can tell you that your mean reversion strategy may not work anymore?



    Forgive me if my questions are trivial. Still finding my way in learning. 
4

1) What is the flaw of my assumption?
 

If there is cointegration then you would get a successful mean reversion strategy. However, the reverse is not true not all successful mean reversion strategy requires cointegration.
 

2) How are professional traders monitor their mean-reversion trading strategy to ensure the rationale behind is right? I mean, what signal can tell you that your mean reversion strategy may not work anymore?

A) A break in cointegration. However, cointegration is not stable estimator, it goes and comes back again. This behaviour is common in stocks pairs. Rather than daily, can calculate at the start of every month or every 15 days.
B) Change in fundamentals of the pair. For example, a pair of two companies where one gets acquired by a third company.
C) Deviation from mean not seen in the backseat. For example, if your pair has never reached a standard deviation of 3 from the mean in the backtest and in live trading, it hits 3.1 then it is a sign to exit and not trade in these pair.

Also, it is advisable to have a portfolio of many pairs, so that loss from a particular pair has a low impact on your capital.

Thanks
 

5

Thousands thanks Ishan! Your answers are very informative and inspiring!

6

Thank you, Bernard. :slight_smile: