Kelly Risk management | High-logical-Best Possible-risk to take so that we can quickly appreciate or depreciate the account balance

Avatar
Belpu Prajwal Rao
  • updated
  • Under review

The Kelly Formula also called the Kelly Criterion, is a very different

risk management method. It is designed to maximize profit but takes much

larger risks than the fixed fractional method.

Kelly grows profits geometrically rather than arithmetically.

***A geometric curve also means that the returns on a portfolio of trades are not even. That means that the returns on wins are bigger than losses.

Here is the formula for determining how much to bet:
F = (bp − q)/b
Where F is the percentage of the current bankroll to wager, 

b is the average winner(in base currency) divided by the average loser(in base currency), 

p is the probability of winning, which is the number of winners divided by the total number of trades,
q is the probability of losing, which is simply 1 – p.

So let’s say that we make 100 pips on average when we win and lose 50 pips when we lose. The ratio is 2:1 so we would put in 2 for the b in the formula. P is the percent of total trades that are winners and q is simply 1 minus p

Replies 13
Avatar
0
Belpu Prajwal Rao

The source of this information is in a book written by Mr Courtney D Smith:

The link to it is here:

Amazon.com: How to Make a Living Trading Foreign Exchange: A Guaranteed Income for Life (Wiley Trading Book 413) eBook : Smith, Courtney: Kindle Store

Just wanted to leave this here attributing the technique rediscovered by Courtney Sir

Avatar
0
Hannes Waser

Yes, nice idea.

how do we estimate the probabilities and the averages? From last N trades? 

or last N positions? (A trade is a chain of positions possibly scaled with martingale)

Avatar
0
Belpu Prajwal Rao
Quote from Hannes Waser

Yes, nice idea.

how do we estimate the probabilities and the averages? From last N trades? 

or last N positions? (A trade is a chain of positions possibly scaled with martingale)

The best way is to backtest your trading system for the entire history. The larger the number of trades you see that it is better and converges to a single same number

Avatar
1
Andrey Khatimlianskyi
  • Under review

Hey, thanks for your idea.

You can calculate estimates with Strategy tester and then put them to the formula. Resulting risk should be set into Lot risk.

If this method works, I can add automatic calculation based on history of trades.

Avatar
0
Belpu Prajwal Rao

Yah, that's wonderful and will properly give the risk values and when the technique stops working it will drastically decrease the 'F' or the %risk for the trade dynamically. Better than any EA which does classic risk management. 
Thanks for all the help.

Also, if risk management can dial up or dial down based on our view of markets then it will be great. For example, we now know that the war in Ukraine will drag EURUSD still more down, so when risk management is considered, EA will take machine calculated risk, but here we can enter with more risk as we know the net risk is low even if the stop loss is hit. Very rarely this will happen if it happens. So can you find a way to favour each currency in a currency pair separated by a ',' as a delimiter for buy or sell mode baising, and if the currency is missing then no human bias is there on it and machine calculated risk can be taken??

Avatar
0
Andrey Khatimlianskyi
Quote from Belpu Prajwal Rao

Yah, that's wonderful and will properly give the risk values and when the technique stops working it will drastically decrease the 'F' or the %risk for the trade dynamically. Better than any EA which does classic risk management. 
Thanks for all the help.

Also, if risk management can dial up or dial down based on our view of markets then it will be great. For example, we now know that the war in Ukraine will drag EURUSD still more down, so when risk management is considered, EA will take machine calculated risk, but here we can enter with more risk as we know the net risk is low even if the stop loss is hit. Very rarely this will happen if it happens. So can you find a way to favour each currency in a currency pair separated by a ',' as a delimiter for buy or sell mode baising, and if the currency is missing then no human bias is there on it and machine calculated risk can be taken??

I'm just not sure if it's worth it.

Have you modeled some results using this method? Share your results with community, please.

Avatar
0
Belpu Prajwal Rao
Quote from Andrey Khatimlianskyi

I'm just not sure if it's worth it.

Have you modeled some results using this method? Share your results with community, please.

I have traded not using the full kelly, (there is something called fractional Kelly which I use) as it takes huge chunks of risks and focuses on real trading, I can tell you that it is dynamic lot allocation +  Larry Williams used it to grow his account with 10000+% in a year of competition using this method...  I don't have proof as I just got started with these techniques and as I was not able to follow the rules very strictly, I started to research creating EA and came across this wonderful project, the CommunityEA and so wanted to use this and implement an EA and optimize it part by part and automate it for a lifetime.
*Check the link for the 11k% on his website
I believe in it and this risk method should not take you more time to implement as a separate function as it is very easy to add and a little old historical analysis with the trading technique to optimize its parameters.

Please take your decision because you are doing the programming. I'm okay with the current style as well. The best would be if you can at least do it in the near future, that will be great.

Avatar
1
drago babnik

The idea is certainly interesting. However, since we are operating here with probability and high risk, the key is how to estimate the b and p coefficients. I think that we can do this reliably enough only with a "forward test" of at least 1/4 of the custom period or at least 3 months of the forward period.

Avatar
0
Belpu Prajwal Rao
Quote from drago babnik

The idea is certainly interesting. However, since we are operating here with probability and high risk, the key is how to estimate the b and p coefficients. I think that we can do this reliably enough only with a "forward test" of at least 1/4 of the custom period or at least 3 months of the forward period.

Thanks for the kind words. Normally, if you follow strict rules based on not less than 3* and not more than 6** independent factors then the law of large numbers says that you will see the trade winning ratio and avg. profits converge to a single number over a large number of trades taken. But yeah, I've seen EA which was profitable using methods which are now not so profitable.
* The 3 & 6 (* &**) factor numbers discussed here, I don't know much about and this can be taken as rumours and so when you supply this information to others please don't tell these as absolute as I don't know why this is true or why not.
Example for factors:

RSI-> Indicates energy and momentum( 50 is lazy or accumulating markets)
OBV-> Activity or Energy in action (mostly indicates where we are headed to)
ADX-> What kind of people are involved in the trade/trend (Below 20 normally professionals entering and exiting trades, 20-40 Normal breakeven traders or losing traders, 40+ retailers mostly moving the markets)

Avatar
0
Andrey Khatimlianskyi
Quote from Belpu Prajwal Rao

I have traded not using the full kelly, (there is something called fractional Kelly which I use) as it takes huge chunks of risks and focuses on real trading, I can tell you that it is dynamic lot allocation +  Larry Williams used it to grow his account with 10000+% in a year of competition using this method...  I don't have proof as I just got started with these techniques and as I was not able to follow the rules very strictly, I started to research creating EA and came across this wonderful project, the CommunityEA and so wanted to use this and implement an EA and optimize it part by part and automate it for a lifetime.
*Check the link for the 11k% on his website
I believe in it and this risk method should not take you more time to implement as a separate function as it is very easy to add and a little old historical analysis with the trading technique to optimize its parameters.

Please take your decision because you are doing the programming. I'm okay with the current style as well. The best would be if you can at least do it in the near future, that will be great.

I can't understand what advantage will you have with automatic risk calculation. Do you want to calc risk on last X trades, so it will change over time? Why is it better than calculating on the previous year history of trades and using this risk all next year?