My journey to long term consistency

Dr Safari, I came across an article in 'Automated Trader' (Issue 39, Q1/2016) regarding the use of kelly betting for the financial markets. The magazine is somewhat difficult to get hold of (and it costs £75 per issue!).

Let me know if you want to discuss it. It addresses the issue of unequal win and loss amounts and the likelihood of overbetting when using Kelly.
Yes, very interested!
 
Quick note: in a 1000 runs with 100 trades each it went bust 17 times. With bust defined as the equity going below 1000 and starting equity was 5200.
 
Demo account

Short Avery Denison
300 Shares
Reason of closure: adjusted stop loss got hit
Opening rate: 73.66
Opening time: 5/19/2016 5:06 PM

Close rate: 74.1
Close time: 5/19/2016 5:27 PM

Amount risked: not sure but probably around 160

Net P/L: €‪-117.82‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 30 min chart. Why 30 now? I waited too long to journal it and 15 mins doesnt fit the screen anymore
AveryDenisonChart.jpg




Personal notes
Looking at it now I understand jack **** of this trade. The stop is WAAAAAAYYYY to narrow for the volatility. I must have been in a hurry and made a mistake, I can't explain it otherwise. Not that it would have worked with a wider stop as you can see but thats not the point of course.
 
Demo account

Long Smith & Wesson Holding
700 Shares
Reason of closure: adjusted stop loss got hit
Opening rate: 23.55
Opening time: 5/24/2016 4:53 PM

Close rate: 23.71
Close time: 5/25/2016 3:38 PM

Amount risked: €169,61

Net P/L: €‪99.15‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 15 min chart
SmithWessonHoldingChart.jpg


Personal notes
Taking into account the decline in the organge circle, I had put my stop loss at 66% of that decline as my system dictates. It hit my adjusted stop loss but after went back up again as you can see. Too bad but it could just as well have been a trend reversal and so this trade is a good one.
 
Demo account

Long Dupont
400 Shares
Reason of closure: adjusted stop loss got hit
Opening rate: 68.36
Opening time: 5/25/2016 4:45 PM

Close rate: 68.11
Close time: 5/25/2016 9:40 PM

Amount risked: €157

Net P/L: €‪-89.58‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 15 min chart
DupontChart.jpg


Personal notes
Trend didnt continue, too bad
 
Demo account

Long Owens Corning
200 Shares
Reason of closure: adjusted stop loss got hit
Opening rate: 51.65
Opening time: 5/24/2016 4:22 PM

Close rate: 51.32
Close time: 5/26/2016 4:26 PM

Amount risked: €161.54

Net P/L: €‪-60.60‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 15 min chart
OwensCorningChart.jpg

Personal notes
Hmmm, the stock went up nice, almost enough to put it to break even but not quite. I'll have to go over my latest journal entries but I think its becoming a pattern of letting winners turn out into losers, maybe I should use 66% profit as the margin of putting break even and not 75%.

Also the stop seems a bit wide for the volatility. Did I make a mistake again? Hmm to investigate further in the future I am also going to mention the price at which I put my stop loss instead of just the amount I am risking

Okay I double checked and it was a correct SL, its just a bit misleading on this chart, bit of optical illusion
 
Quick note: in a 1000 runs with 100 trades each it went bust 17 times. With bust defined as the equity going below 1000 and starting equity was 5200.

Thanks for sharing this info.

Starting Equity: 5,200
Bet Size: 26% of remaining capital
"Bankruptcy" Level: 1,000

Are those the parameters you used for that last run?

Are you able to display the simulation results in a scattergraph format? That would be super!

The quick answer as to how to treat unequal wins & losses - (I can't refer to the article right now, because my colleague borrowed my issue over the bank holiday weekend - so I am going by memory) - basically you take into account the standard deviation/variance and the skew (if any) of your results in order to calculate the likelihood of overbetting using Kelly - whether full Kelly or fractional.

The skew is negative if there are some large negative outliers . The skew is positive if there are some large positive outliers (apologies if that's already as clear as mud to you!!). If the results are normally distributed, with the mean being equal to your expectancy/trade, then there is no skew. I would assume that your results are positively skewed as you have a fixed stop loss in terms of $, and also a few large winning trades. If I remember correctly, overbetting with Kelly is more likely to occur for data that is negatively skewed.

We can discuss this in a bit more detail still, but I suggest we do that via PM - as it might get pretty detailed.
 
Thanks for sharing this info.

Starting Equity: 5,200
Bet Size: 26% of remaining capital
"Bankruptcy" Level: 1,000

Are those the parameters you used for that last run?
Correct!

Are you able to display the simulation results in a scattergraph format? That would be super!
Is this what you had in mind? X-axis is run 1 to 1000, y-axis the equity result.

100 Trades per run:
scatterGraphSimulation100Trades.jpg

50 Trades per run:
scatterGraphSimulation50Trades.jpg



basically you take into account the standard deviation/variance and the skew (if any) of your results in order to calculate the likelihood of overbetting using Kelly - whether full Kelly or fractional.

The skew is negative if there are some large negative outliers . The skew is positive if there are some large positive outliers (apologies if that's already as clear as mud to you!!). If the results are normally distributed, with the mean being equal to your expectancy/trade, then there is no skew. I would assume that your results are positively skewed as you have a fixed stop loss in terms of $, and also a few large winning trades. If I remember correctly, overbetting with Kelly is more likely to occur for data that is negatively skewed.
Yeah I get what you are saying but to actually calculate it (and being able to implement it into the simulator) I'm afraid I'll have to brush up on my statistics knowledge :p

We can discuss this in a bit more detail still, but I suggest we do that via PM - as it might get pretty detailed.
Either way is fine by me, it might clutter this journal a bit too much indeed.
 
Last edited:
Safari!

If you didnt already, you should read some of Ralph Vince's books. Probably "mathematics of money management" will do. It goes into more detail on (fractional) kelly. Also discusses other distributions than the normal (quite useful). So you can also brush your statistics up a bit ;)

Kind regards,

Koen
 
Safari!

If you didnt already, you should read some of Ralph Vince's books. Probably "mathematics of money management" will do. It goes into more detail on (fractional) kelly. Also discusses other distributions than the normal (quite useful). So you can also brush your statistics up a bit ;)

Kind regards,

Koen

Thx it's on my to read list!
 
My latest demo trades havent been working out. I have gone over my journal and I believe the main reason for this is not selecting the right stocks, most of what I defined in my system about money and risk management and market sentiment and timing etc are still valid except the one thing I am not strict enough on is this:

- scan american stocks and look for clean charts (whether up or down determinded in previous point). Clean is a rising or falling 10 period WMA with no interruptions for at least 3 trading days

I gave a definition of a clean chart but its not very good as it allows me to select not so clean charts meaning having too much volatility and these trades tend to not work out so well most of the time. So now I am going to adapt my system once more and define how much max volatility is allowed in a stock.

I am first reading this and then post my new system in a new post: http://www.tradeciety.com/how-to-use-the-atr-indicator-guide/ (not sure if that links is allowed but if not feel free to remove it)

Damn Im facing the same problem I had before: the ATR numbers are always different so its difficult to define
 
Last edited:
Okay I am not going to give a clear measurable definition of a clean chart but a very clear description of what I am looking for. I dont like it but it will have to do for now. I am also making a few minor adjustments:

Plan Demo Account including new findings:

1) Deciding whether to trade or not
-before deciding to trade look if any major economic events are up in the current trading day if yes then we DO NOT TRADE unless after 1 hour of occurence
-before deciding to trade: check yahoo finances whether something big is up or not. If it is DO NOT TRADE


2) Find potential trades
- Wait at LEAST 1 hour after trading has opened before taking up a position
- Wait at LEAST 1 hour after the occurence of a major economical event or indicator (1)
- Look at the S&P500 and the stocks in your watchlist. Watch if most of them are in the plus or in the min. Dont go long if in the min and dont go short if in the plus.
- scan american stocks and look for clean charts (whether up or down determinded in previous point). Clean will be "defined" below (2)
- do not include stocks that gapped opun opening of that day. Gap meaning an obvious white space between two candles on the 15 minute chart
- do not include stocks that require me to take stop losses wider than 2% of my equity
- the last three bars (1h) should go with the trendline, else dont take it



3) Taking positions
- look for the trend and 66% of the highest difference in opposite direction of the entire trend is where I place my stop loss
- I am not going to risk more than 2 % on each trade



4) Once in a position
- Make it a trailing stop loss
- Once the trade reaches 75% profit of amount risked move the SL to break even.
- Once the trade reaches 100% profit of amount risked move the SL to 50% profit (of what was the SL)

- Exit only if stop loss hits
- document every trade carefully in journal (with charts) saying what you did and what I should have done

Since the system changed again I will evaluate after 40 trades (or sooner if it seems totally worthless like I did many times already)

(1) If this is the case I am allowed to look at the 15min charts to check whether trend continues or not

(2) A clean chart is a chart with:
- As much candles as possible in the same colour.
- As little wicks as possible
- All candles having about the same height.
- The top of the previous candle comes close to the bottom of the next candle (or vice versa)




STILL TODO: find out how the trailing SL works in demo account
 
I made some additional changes:



Plan Demo Account:

1) Deciding whether to trade or not
-before deciding to trade look if any major economic events are up in the current trading day if yes then we DO NOT TRADE unless after 1 hour of occurence
-before deciding to trade: check yahoo finances whether something big is up or not. If it is DO NOT TRADE


2) Find potential trades
- Do this (the following) before US market opens and make a list of potential candidates
- One hour after opening of market go through that list and look for positions to take

- Wait at LEAST 1 hour after the occurence of a major economical event or indicator (1)
- Look at the S&P500 and the stocks in your watchlist. Watch if most of them are in the plus or in the min. Only go Long if the market is clearly more bullish, only go short when the market is clearly more bearish, do nothing if that is not clear.
- scan american stocks and look for clean charts (whether up or down determinded in previous point). Clean will be "defined" below (2)
- do not include stocks that gapped opun opening of that day. Gap meaning an obvious white space between two candles on the 15 minute chart
- do not include stocks that require me to take stop losses wider than 2% of my equity
- the last three bars (1h) should go with the trendline, else dont take it



3) Taking positions
- look for the trend and highest difference in opposite direction of the entire trend is where I place my stop loss
- I am not going to risk more than 2 % on each trade
- I am not going to risk more than 6 % in total positions



4) Once in a position
- Make it a trailing stop loss
- Once the trade reaches 75% profit of amount risked move the SL to break even.
- Once the trade reaches 100% profit of amount risked move the SL to 50% profit (of what was the SL)
- Exit only if stop loss hits
- document every trade carefully in journal (with charts) saying what you did and what I should have done

(1) If this is the case I am allowed to look at the 15min charts to check whether trend continues or not

(2) A clean chart is a chart with:
- As much candles as possible in the same colour.
- As little wicks as possible
- All candles having about the same height.
- The top of the previous candle comes close to the bottom of the next candle (or vice versa)


Since the system changed again I will evaluate after 40 trades (or sooner if it seems totally worthless like I did many times already)
 
Last edited:
I made some additional changes:

Great stuff, Dr Safari!

I know what you mean by having difficulties in defining a "clean chart". I have similar issues in assessing whether price is 'respecting' my moving averages, and whether 'momentum' has occurred. At one point, I made a list of the various factors that I was looking at in making my judgement (to the extent I was consciously aware of these factors) - I then used these factors as a checklist. Over time, and looking at hundreds of setups, my brain can now start to spot whether a pattern is valid or not. I can't always fully explain why a setup is valid or not, but after a while it's just a case of looking at it and going 'yes' or 'no'. And if it's a 'maybe', then I don't take it.

IMHO, using our brains in this manner is one advantage that we have over algos in that it might be difficult to program these human brain processes. At least more difficult than scanning through dozens of black&white checks in order to execute a trade in a millisecond.

Good luck with your ongoing process.

PS Will shortly PM re those Kelly details.
 
Over time, and looking at hundreds of setups, my brain can now start to spot whether a pattern is valid or not. I can't always fully explain why a setup is valid or not, but after a while it's just a case of looking at it and going 'yes' or 'no'. And if it's a 'maybe', then I don't take it.

IMHO, using our brains in this manner is one advantage that we have over algos in that it might be difficult to program these human brain processes. At least more difficult than scanning through dozens of black&white checks in order to execute a trade in a millisecond.
Yeah fully agreed! I recently read "Blink" by Malcolm Gladwell and it's about exactely that phenomenon. That experts in their area of expertise unconsciously feel when something is right or wrong but can't necessarily explain why. Unconscious learning is a real thing and its helping us a lot and I agree that that is THE most important advantage we still have over trading algorithms... or any algorithms for that matter.

One thing concerns me though. I'm not sure if you are familiar with the eastern/asian game that's considered the asian rival of chess? It's called "Go". What's so special about it is that computers haven't been able to beat the best players... until recently: http://www.wired.com/2016/01/in-a-huge-breakthrough-googles-ai-beats-a-top-player-at-the-game-of-go/. Brute force calculations also don't help much to win this game because the amount of possible configurations of the game exceeds the number of atoms in the universe. Google AI made an AI that could learn from reviewing thousands of games and then further learn by letting it play games against itself.

If they do the same for trading.... Though trading is always more complex than a game with only a few basic rules of course. We will see I guess

Good luck with your ongoing process.
Thanks! :)
 
Demo account

Long Atlassian
550 Shares
Reason of closure: stop loss got hit
Opening rate: 24.98
Opening time: 6/2/2016 4:38 PM

Close rate: 24.73
Close time: 6/2/2016 5:03 PM

Amount risked: €161.54
Stop loss: 24.73

Net P/L: €‪-123.19‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 15 min chart
AtlassianChart.jpg

Personal notes
As you can see, the new system allows for a very narrow stop loss. Which also allows me to take relatively big positions, which will (hopefully) turn out in bigger profits too when the trade does work.
 
Demo account

Long Assured Guaranty
1250 Shares
Reason of closure: stop loss got hit
Opening rate: 27.4
Opening time: 6/2/2016 4:34 PM

Close rate: 27.28
Close time: 6/2/2016 5:35 PM

Amount risked: €‪134.36
Stop loss: 24.28

Net P/L: €‪-134.43‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 15 min chart
AssuredGuarantyChart.JPG

Personal notes
Trade did not work. I made three trades in total, 2 of which are now stopped out and already documented here but a third one is still running and already compensating for the loss of these two. Time will tell if my new way of doing things will be profitable over time. After that I can do the same on real account with small positions, then increase size according to kelly criterion... and hopefully make a lot of money so I can do this for a living :D ... but first things first


Maybe I should only post the hourly charts since thats what I base myself on when scanning??? Then again not much to see there then, it's already cutting it close with these 15 min charts. Posting both would be ideal but time consuming.
 
Demo account

Long Aetna
150 Shares
Reason of closure: adjusted stop loss got hit
Opening rate: 117.35
Opening time: 6/2/2016 4:31 PM

Close rate: 118.78
Close time: 6/3/2016 3:34 PM

Amount risked: €‪139.71
Stop loss: 116.31

Net P/L: €‪+188.28‬


Chart
Green arrow is where I opened position and red arrow is where it got closed. Chart is a 15 min chart
AetnaChart.jpg

Personal notes
This was initially 350 in the plus, but upon opening it gapped down and hit my adjusted SL. I knew this was going to happen cause I was watching the S&P500 futures. Because of that I put my SL even higher but of course that didnt help. This is the disadvantage of holding things overnight, then again you have to let your profits run too.

It gapped down because of the unemployment rate economic event. I could integrate it into my strategy to get out of my positions when I still can when a high expected volatility economic event is going to occur . Hmmm this does not fit badly with my renewed strategy actually. Previously it wouldnt have worked as well because it was unlikely to make enough profit during one day, now its perfectly possible.

Okay lets make it a rule. I will add it to the plan.
 
As promised yet another revised edition

Plan Demo Account:

1) Deciding whether to trade or not
-before deciding to trade look if any major economic events are up in the current trading day if yes then we DO NOT TRADE unless after 1 hour of occurence
-before deciding to trade: check yahoo finances whether something big is up or not. If it is DO NOT TRADE


2) Find potential trades
- Do this (the following) before US market opens and make a list of potential candidates
- One hour after opening of market go through that list and look for positions to take
- Wait at LEAST 1 hour after the occurence of a major economical event or indicator (1)
- Look at the S&P500 and the stocks in your watchlist. Watch if most of them are in the plus or in the min. Only go Long if the market is clearly more bullish, only go short when the market is clearly more bearish, do nothing if that is not clear.
- scan american stocks and look for clean charts (whether up or down determinded in previous point). Clean will be "defined" below (2)
- do not include stocks that gapped opun opening of that day. Gap meaning an obvious white space between two candles on the 15 minute chart
- do not include stocks that require me to take stop losses wider than 2% of my equity
- the last three bars (1h) should go with the trendline, else dont take it



3) Taking positions
- look for the trend and highest difference in opposite direction of the entire trend is where I place my stop loss
- I am not going to risk more than 2 % on each trade
- I am not going to risk more than 6 % in total positions



4) Once in a position
- Make it a trailing stop loss
- Once the trade reaches 75% profit of amount risked move the SL to break even.
- Once the trade reaches 100% profit of amount risked move the SL to 50% profit (of what was the SL)
- Exit only if stop loss hits OR if a high expected volatility economic event is going to occur
- document every trade carefully in journal (with charts) saying what you did and what I should have done

(1) If this is the case I am allowed to look at the 15min charts to check whether trend continues or not

(2) A clean chart is a chart with:
- As much candles as possible in the same colour.
- As little wicks as possible
- All candles having about the same height.
- The top of the previous candle comes close to the bottom of the next candle (or vice versa)


I also investigated the trailing SL possibility and it seemed to work fine now so using that from now on.
 
Top