Im realizing something truly disappointing with using a hard sl with the mean reversion strategy


Im using entry at 1 Stddev and 1.5 Stddev SL with the moving average as the take profit I notice that my hard stop loss which makes my risk to reward as RR 1:2 doesnt mean my assets will take a max loss of 50% of my winnings.
Does this seem to be normal? Should using a hard stop loss with RR of 1:2 with my z score protect me some loss but DOESNT MEAN on the real asset ill be risking 50% of my usual reward?


Thanks again.
 

Hello Jane, 

The risk/reward itself can't give you the complete picture. It would be best if you also considered the hit rate. 



For your mean reversion strategy with entries/exits as per standard deviation, you can improve the hit rate by trying other combinations of entry and exits like (1.5, 2),(2, 2.5),(3,2.5) etc to find the combination of risk/reward with better hit rate.

 

In general, many trading systems use risk/reward better than 1/2 say 1/3 i.e. risking $1 to gain $3.

Hope this helps! please leave a comment if you have any further questions on this.



 

What would be the best way to get a hard sl for a mean reversion system? Not the spread or zscore but the 2 assets having a hard stop loss this is possible or advisable?

Hello Jane, 

Theoretically, the stop loss is the price point where your entry logic fails.

For this reason, in most cases, the stop loss is a function of the variables that generated the entry signal. Hence, the stop loss of the mean reversion system you are referring to depends on zscore. However, it's not uncommon to have a risk-based stop loss depending on the risk allocated per trade as a hard stop loss. 

However, it's always advisable to give preference to the zscore based stop loss.



Hope this helps!

Im looking for papers with the hard SL on the both assets, it would be perfect in my opinion to optimize this. Is there any or anyone who may know how to get this or would know about this please let me know or email me.



Thank you.

Namaste.

Hello Jane, here are some papers on optimising risk management of pairs trading.


  1. Optimal Mean Reversion Trading with Transaction Costs and Stop-Loss Exit - Tim Leung, Xin Li (International Journal of Theoretical and Applied Finance, Vol. 18, No. 3, 2015) link


  2. On the Efficacy of Optimized Exit Rule for Mean Reversion Trading -  Donovan Lee, 

    Tim Leung link


  3. Optimal liquidation of a pairs trade - Ekstrom, E., Lindberg, C., and Tysk, J. (2011) link



    I hope this helps! 

Is it possible to use pending orders with a mean reversion system?

Hello Jane, 

Yes, it is possible to use pending orders with a mean reversion trading system.



In a mean reversion trading system, pending orders can be used to enter trades when the price of the security moves away from the mean and is expected to revert back towards it.



For example, if a mean reversion system has identified a security as being oversold, a trader could use a buy limit order to enter a long position when the price falls to a level that is considered attractive to open a long position.



Similarly, if a security is identified as being overbought, a trader could use a sell-stop order to enter a short position when the price rises to a level that is considered attractive to open a short position.



Pending orders can be useful in a mean reversion system because they allow the trader to set the entry price in advance, rather than having to manually monitor the market and place the trade manually when the price reaches the desired level. 



On the other hand, pending orders can't be used for the pairs trading system since it has the risk of executing an order in only one of the pairs and other order might not get executed. And also, with a pending order, there is a change that only partial order quantity might get executed. This makes it impossible to maintain the hedge ratio.



I hope this helps!

What did you mean by this? "However, it's not uncommon to have a risk-based stop loss depending on the risk allocated per trade as a hard stop loss."



If you have any examples or papers i would appreciate it.



Namaste.

Hello Jane, 

I was explaning the 'Risk-per-trade approach'. 



This method revolves around protecting your own capital. As far as stop loss strategy options go, the risk-per-trade approach to reduce increasing losses is simple and is regarded as an effective way to manage risk when trading. 



By determining the maximum amount of risk you are willing to take on each trade, you can set a stop loss that will limit your losses if the trade goes against you. 



The maximum risk you are willing to accept on a trade should be a dollar amount, or whatever currency your trading account is based in. The dollar amount is determined as a percentage of the total value of the trading capital in your account. 



There are a few different ways to calculate your risk-per-trade, but one of the most popular is the 2% rule. This principle says that you should never risk more than 2% of your account on any single trade. So, if you have a $10,000 account, you should never risk more than $200 on a trade.



You can check these links for further reading (link1, link2)



I hope this clarifies your doubts. 

So for a mean revrsion pair trading system you would use the underlying support and resistance of each asset as max hard stop loss lines? So 4 lines in total per pair?

Yes Jane, thats right.

Thanks again! I forget to say thanks.





Namaste.

Let’s say I have lookback and the number of extremas as variables. How would you optimize the features of the function that creates the support and resistance lines? (grid search random search, bayes)

Hi Jane, 

Since you have already decided on the hyperparameters, to optimize the hyperparameters of the function that creates the support and resistance lines, one should define a metric to evaluate the performance/accuracy of the function.



Once this is done, you can use any of the optimisation methods you have mentioned to study the performance/accuracy of each combination of the hyperparameters. 



You can also compare the results obtained with different optimization techniques (e.g., grid search, random search, Bayesian optimization) to identify the best-performing combination of hyperparameters. 



I hope this helps!

Would you use bayes and DRL? or just use Deep RL to do them all?