Kelly Criterion - Optimizing risk | Page 2 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Using the Kelly Criterion in actual trading 18 replies

Optimizing "The Sauce" 14 replies

Need help optimizing custom indicator 7 replies

Optimizing expert advisors with metatrader 4.00 0 replies

Thoughts on backtesting and optimizing 6 replies

Attachments: Kelly Criterion - Optimizing risk
Exit Attachments

Kelly Criterion - Optimizing risk

  • #21
  • Jan 12, 2017 4:43am Jan 12, 2017 4:43am
Discovered some hidden errors in the spreadsheets. Again. Will this ever end? I hope so. But the main part (exits) is working fine. Balancing risk with performance is a tricky task. Still not where it should be, but I will get there. Kelly is very aggressive. Looks unrealistic with the data available. Pushing 10% of suggested figure for now. Hmmm, this thread is looking like a journal. A lonely journal, lol. Useless? Maybe. Some errors are not corrected yet (e.g. Payoff looks wrong to me). I am trying to figure out how to fix it.
Attached File(s)
File Type: xlsx Toolbox.xlsx   209 KB | 528 downloads
File Type: xlsx Monte Carlo Simulation.xlsx   25 KB | 558 downloads
  • #22
  • Jan 24, 2017 4:13am Jan 24, 2017 4:13am
Attached here is the best simple explanation of Kelly Criterion I have seen so far.
Attached File(s)
File Type: pdf Kelly Formula Derivation.pdf   67 KB | 1,060 downloads
  • #23
  • Jan 25, 2017 4:36pm Jan 25, 2017 4:36pm
Many improvements were made to the "toolkit", one affecting the Monte Carlo spreadsheet, updated. Main payoff figure is now being estimated corretly. Kelly % is definitely a fantasy. Maximum leverage implemented. Next step: reduce stop out to 25%. Trying to implement some sort of stop out x payoff analysis.
Attached File(s)
File Type: xlsx Toolbox.xlsx   229 KB | 486 downloads
File Type: xlsx Monte Carlo Simulation.xlsx   25 KB | 593 downloads
  • #24
  • Jan 26, 2017 3:01pm Jan 26, 2017 3:01pm
The simpler the better. A chart now shows the evolution of payoff with increased leverage (and decreased stop out). At some point a balance should be reached (shall it?). The question is: "Which leverage should I use to get maximum payoff - if any?" Shame on me I didn't do it from start.
Attached File(s)
File Type: xlsx Toolbox.xlsx   233 KB | 515 downloads
  • #25
  • Edited 1:20pm Jan 30, 2017 12:27pm | Edited 1:20pm
Market pays tithing for hell rides (each 1% drawdown is returning close to 0.1% extra on average). Hope it doesn't let me head down stuck in the 9th circle someday, lol. "Tithe chart" improved (unnamed scatter chart on DDCy sheet). Cycle, daily, weekly, monthly and yearly estimated payoffs added (the latter with a timeline chart). Did not lower stop out yet (waiting for a more realistic probability of hitting it). Holding at 30%.
Attached File(s)
File Type: xlsx Toolbox.xlsx   235 KB | 485 downloads
  • #26
  • Feb 4, 2017 4:03pm Feb 4, 2017 4:03pm
Many improvements were made to the risk balance mechanism. Estimated probabilities look realistic now. Got rid of some fancy numbers.
Attached File(s)
File Type: xlsx Toolbox.xlsx   262 KB | 514 downloads
  • #27
  • Feb 9, 2017 6:02pm Feb 9, 2017 6:02pm
The risk balance mechanism is working as it should. Overshooting a bit maybe, but this may be compensated by the "aggressiveness" variable. There was a copy/paste error in some cells, now corrected. The spreadsheet is fully operational. Three unimportant payoff figures still to be fixed someday.
Attached File(s)
File Type: xlsx Toolbox.xlsx   260 KB | 801 downloads
  • #28
  • Jun 12, 2018 2:11am Jun 12, 2018 2:11am
Quoting Jairo
The risk balance mechanism is working as it should. Overshooting a bit maybe, but this may be compensated by the "aggressiveness" variable. There was a copy/paste error in some cells, now corrected. The spreadsheet is fully operational. Three unimportant payoff figures still to be fixed someday. {file}
Oi Jairo! Tudo bom?

I found this thread whilst looking for Excel adaptations of the Kelly Criterion. You've done an incredible job, congratulations and thank you for sharing all of this amazing data.

I would like to get in touch with you regarding the sheet, if you can spare a couple of minutes to do so. Cheers!
  • #29
  • Jun 25, 2018 7:12am Jun 25, 2018 7:12am
Quoting elqflinchy
{quote} Oi Jairo! Tudo bom? I found this thread whilst looking for Excel adaptations of the Kelly Criterion. You've done an incredible job, congratulations and thank you for sharing all of this amazing data. I would like to get in touch with you regarding the sheet, if you can spare a couple of minutes to do so. Cheers!
Hi elqflinchy,
I stopped using double down (martingale) and cleaned the spreadsheet from it. Updated files are attached (English/Portuguese mixed, lol). Results still a bit disappointing, but Kelly with low "aggressiveness" keeps me from overleveraging.
Attached File(s)
File Type: xlsx Monte Carlo Simulation.xlsx   113 KB | 793 downloads
File Type: xlsx Operações.xlsx   162 KB | 716 downloads
File Type: xlsx Resultados.xlsx   152 KB | 681 downloads
  • #30
  • Aug 13, 2020 7:30am Aug 13, 2020 7:30am
This is the apreadsheet I am using now - for whatever it is worth. It is still a work in progress but it has been useful for me. Trading guidelines are explained in the Rules tab. Disclaimers: (1) Kelly is just a reference - too shaky to be used in full; (2) my loss limiting policy is still under test.
Attached File(s)
File Type: xls Gerenciamento de risco.xls   390 KB | 636 downloads
  • #31
  • Aug 15, 2020 7:12am Aug 15, 2020 7:12am
Nice thread, great work.

I think one thing worth mentioning is that Kelly is a little aggressive for most people and often people will trade at half or quarter Kelly. I've tried to explain it clearly below to add to your thread

It is designed to long-term optimise bet size in games with predictable odds but it assumes that you can take big drawdowns!

This is harder than it sounds. Let’s say you could bet on a weighted coin flip, where it lands on heads 60% of the time and tails 40% of the time. The payout is $2 per $1 bet.

  1. Well, absolutely you should bet. The odds are in your favour. But if you have, say, $100 it is less obvious how much you should bet to avoid ruin.
  2. Say you bet $50, the odds that it could land on tails twice in a row are 16%. You could easily be out after the first two flips.
  3. Equally, betting $1 is not going to maximise your advantage. The odds are 60/40 in your favour so only betting $1 is likely too conservative. The Kelly Criterion is a formula that produces the long-run optimal bet size, given the odds.


Applying the formula to forex trading looks like this:

Position size % = Winning trade % - ( (1- Winning trade %) / Risk-reward ratio )

If you have recorded hundreds of trades in your journal - see next chapter - you can calculate what this outputs for you specifically. If you don't have hundreds of trades then let’s assume some realistic defaults of Winning trade % being 30% and Risk-reward ratio being 3. The 3 implies your TP is 3x the distance of your stop from entry e.g. 300 pips take profit and 100 pips stop loss.

So that’s 0.3 - (1 - 0.3) / 3 = 6.6%.

Hold on a second. 6.6% of your account probably feels like a LOT to risk per trade.

This is the main observation people have on Kelly: whilst it may optimise the long-run results it doesn’t take into account the pain of drawdowns. It is better thought of as the rational maximum limit. You needn’t go right up to the limit!

With a 30% winning trade ratio, the odds of you losing on four trades in a row is nearly one in four. That would result in a drawdown of nearly a quarter of your starting account balance. Could you really stomach that and put on the fifth trade, cool as ice? Most of us could not.

Accordingly people tend to reduce the bet size. For example, let’s say you know you would feel emotionally affected by losing 25% of your account.

Well, the simplest way is to divide the Kelly output by four. You have effectively hidden 75% of your account balance from Kelly and it is now optimised to avoid a total wipeout of just the 25% it can see.

This gives 6.6% / 4 = 1.65%. Of course different trading approaches and different risk appetites will provide different optimal bet sizes but as a rule of thumb something between 1-2% is appropriate for the style and risk appetite of most retail traders.

  • #32
  • Aug 15, 2020 4:39pm Aug 15, 2020 4:39pm
Quoting getmrmarket
Nice thread, great work. I think one thing worth mentioning is that Kelly is a little aggressive for most people and often people will trade at half or quarter Kelly. I've tried to explain it clearly below to add to your thread It is designed to long-term optimise bet size in games with predictable odds but it assumes that you can take big drawdowns! This is harder than it sounds. Let’s say you could bet on a weighted coin flip, where it lands on heads 60% of the time and tails 40% of the time. The payout is $2 per $1 bet. Well, absolutely...
Hi getmrmarket,
Thanks for your great contribution. I see you have much better math background than mine. Maybe I can abuse a little and ask you a question that is puzzling me at the moment. I calculate reward to risk ratio after every trade considering "risk" to be the biggest drawdown experienced during the trade and taking the final profit as "reward". Then I keep track of the average R/R along the series of trades. What I can't understand is why the arithmetic mean looks much "better" than the geometric mean, as you can see in the "Reward to Risk" tab charts. I would expect the contrary, as percent profits and losses are compounded in the account history, not just added. Please disregard my stupid question but I am still - and always - learning as I try to play this extremely challenging game called trading.
Best regards.
  • #33
  • Aug 16, 2020 3:43pm Aug 16, 2020 3:43pm
I use Half Kelly Criterion - I didn't know traders on FF were also using it.

I use this calculator from Albion Research :

https://www.albionresearch.com/kelly/default.php

I'm risking 5% a trade at Half Kelly Criterion.
All Assets NAS100 GLD BTC AUS JPY GBP All Time Return: 17.5%
  • #34
  • Aug 21, 2020 3:21am Aug 21, 2020 3:21am
Very good work you have done here sir. Could you possibly translate the Excel workbooks to English please? Thanks in advance
Dig your ditches. Dont fill them in
  • #35
  • Aug 21, 2020 9:13pm Aug 21, 2020 9:13pm
Quoting Irpheus
Very good work you have done here sir. Could you possibly translate the Excel workbooks to English please? Thanks in advance
Thank you for your kind words. It may take some time (it is not so easy for me). The main trade log tabs are in English. I did a quick copy/paste test of the rules into Google translator and it didn't look bad. You may ask/criticize anything you want here. I am a brutal critic of everything I do. I am not sure this method is worth the effort. There may be changes, improvements, some guidelines may be proven wrong with time...
  • #36
  • Edited Jun 7, 2021 11:55am Jun 6, 2021 6:52pm | Edited Jun 7, 2021 11:55am
Hi,

I'm doing experiments on Kelly criterion trying to find something useful.

On the attachment there is a modified version of an excel from author: Peter Ponzo [In memory of], aka Gummy.

The example is applied to BTC-USD. [You could experiment with any Stock history]

There are 2 Kelly Formulas:

-Regular Kelly [ignore the volatility and assume constant daily returns equal to the average.]

x = p - q/(W/L)

-gk Kelly [mimics the volatility and assume constant daily returns equal to the average.]
x = (p*W - q*L) / {p*S2(W)+p*W2 + q*S2(L)+q*L2}

On each Kelly formula you could modified Portfolio size [default $1,000] and Kelly% [default 100%].

https://i.ibb.co/x8G8xrt/GLK.png


Regards
Attached File(s)
File Type: xlsx GLS KELLY.xlsx   240 KB | 351 downloads
  • #37
  • Jun 9, 2021 6:52am Jun 9, 2021 6:52am
Quoting HyperD
Hi, I'm doing experiments on Kelly criterion trying to find something useful. On the attachment there is a modified version of an excel from author: Peter Ponzo [In memory of], aka Gummy. The example is applied to BTC-USD. [You could experiment with any Stock history] There are 2 Kelly Formulas: -Regular Kelly [ignore the volatility and assume constant daily returns equal to the average.] x = p - q/(W/L) -gk Kelly [mimics the volatility and assume constant daily returns equal to the average.] x = (p*W - q*L) / {p*S2(W)+p*W2 + q*S2(L)+q*L2} On each...
Hi, HyperD,
You did a great job with your Kelly spreadsheet. Thank you for sharing.
  • #38
  • Jun 9, 2021 9:21am Jun 9, 2021 9:21am
Quoting Jairo
{quote} Hi, HyperD, You did a great job with your Kelly spreadsheet. Thank you for sharing.
Hi Jairo,

I can't find on web search Kelly formulas that include volatility, that's the main reason I'm sharing Peter Ponzo [RIP] algorithm, to compare both scenarios.

Sometimes gk Kelly suggests more than 100% the X% [leverage greater than 1:1?].

Regular Kelly EAs are already available. I don't know if an EA with gK Kelly alghorithm could be done.

NOTE: X% stop loss should be included in real trading.


Regards
  • #39
  • Jun 9, 2021 10:44am Jun 9, 2021 10:44am
Quoting HyperD
{quote} Hi Jairo, I can't find on web search Kelly formulas that include volatility, that's the main reason I'm sharing Peter Ponzo [RIP] algorithm, to compare both scenarios.
HyperD: if you're looking for the Kelly formula that includes volatility, you may be searching for continuous Kelly criterion (second link below). I've posted a few thoughts on this a long time ago, including an adaptation of the Kelly continuous growth rate formula here that could be applicable for traders:

https://www.forexfactory.com/thread/...67#post6961167

And probably of more interest to you, if you follow the link below you'll find an interesting and accessible pdf that shows how to use the continuous Kelly criterion formula to enhance buy and hold as well as a fairly easy to understand derivation of the formula:

https://www.forexfactory.com/thread/...81#post6418881

The chapter in the Wilmott book (referenced in first link) is also a must-read if you wish to understand this subject and formulas inside and out.
  • #40
  • Jun 9, 2021 10:59am Jun 9, 2021 10:59am
Quoting FXEZ
{quote} HyperD: if you're looking for the Kelly formula that includes volatility, you may be searching for continuous Kelly criterion (second link below). I've posted a few thoughts on this a long time ago, including an adaptation of the Kelly continuous growth rate formula here that could be applicable for traders: https://www.forexfactory.com/thread/...67#post6961167 And probably of more interest to you, if you follow the link below you'll find an interesting and accessible pdf that shows how to use the continuous Kelly criterion...

Thanks FXEZ for the links.
0 traders viewing now