Trade Planning and Testing

[Kodiak]

On 11/08/2020 at 15:01, dmedin said:

I did the same thing on Dax, Tom, and got a similar win rate (49%)

This is the running equity position, based on points  It's wonderful!  I'm already trading this system on multiple markets

Sell it to Jim Simons for a million or two🤑

https://markets.businessinsider.com/news/stocks/jim-simons-renaissance-technologies-rentech-apple-amazon-tesla-stock-2020-8-1029504458#

[Caseynotes]

1 hour ago, dmedin said:

The Pine editor in Tradingview is great!  Just tested the simplest strategy on daily Ocado share price: buy when 10 SMA crosses 50 SMA and sell the inverse.  Clicked a button and it ran a backtest and gave me the results: 619.00 net profit from 62 trades, of which about 34% were profitable.   

see the importance of stats, without the full data picture you would have noticed that the win rate was low and just moved on to something else, forever going round in circles.

Edited August 18, 2020 by Caseynotes

[dmedin]

9 minutes ago, Caseynotes said:

see the importance of stats, without the full data picture you would have noticed that the win rate was low and just moved on to something else, forever going round in circles.

 

Yes, it is an absolutely vital element  None of the old trading books mentioned it.

[dmedin]

I've been falling into the trap of trading outside of my 'plan' and it almost always results in losses.

The other thing I realized is that when backtesting I was effectively ignoring the news and events and just going off the indicator.  That means there is really no reason to be at the trading all day long and checking for events as I don't trade them.  I can literally not read any financial or economic information whatsoever and simply place trades based on the indicators.

Some of my backtested trades last between eight days and four weeks.  It's easy to let a backtested trade last for four weeks (and then ultimately make a loss in some cases) than it is to be checking up on it every hour of every day for four weeks live  

Edited August 20, 2020 by dmedin

[dmedin]

Backtesting FX this weekend.  Going to try for risk-to-reward of 1:1.5 instead of 1:2.

I like H4 for forex so that I can capture significant moves at minimum bet size.  But will obviously have to test H1 as well.

[dmedin]

Big improvement in results so far using 1:1.5 instead of 1:2 🤓

[dmedin]

100 backtests of the U.S. dollar index, H4 timeframe with R:R of 1:1.5 gives

53% win rate

25.61 points

But the first trade is in October 2015, and the last trade is dated 3rd August this year! 

And for 2019, the account lost money!  That would take a fair bit of patience to endure in 'real time'!

Time to test hourly ...

And then try optimizing and tweaking things 🤯

Edited August 23, 2020 by dmedin

[dmedin]

Any tips, Tom? 🤓

[dmedin]

There are so many other indicators I have to test.  And other variables to consider for example the slope of the 200 EMA.  And then of course I will eventually have to start writing Pine scripts and backtest shares.  Oh I have many hours of work ahead of me 🤯💨

[Caseynotes]

18 hours ago, dmedin said:

Any tips, Tom? 🤓

yes, starting a new thread would be choice.

you don't want to trade the US basket so don't do a test on it.

your system needs to be practical, if there are too few trades then it needs to work on many multiple markets so forget testing over multiple years and test multiple markets instead.

what does 25.61 points mean? you don't mean you gained 25 and a half points on 50 winning trades.

[dmedin]

2 hours ago, Caseynotes said:

your system needs to be practical, if there are too few trades then it needs to work on many multiple markets so forget testing over multiple years and test multiple markets instead.

I've done 50 trades on EUR/USD so far, hourly with 1:1.5 and it's profitable, and will do all 100 and then try other pairs.

With forex, if betting £1 per point then it doesn't really justify the effort ... would need to trade multiple pairs as you say or bet more pounds per point 🤑

1:1.5 works better with forex so far, as it doesn't trend as nicely as indices do

[dmedin]

I might also try trading candle reversals off pivot points 100 times 🤑

[dmedin]

Will need to try 1:2 risk-to-reward with EUR/USD as seem to be missing out on profits when getting it right  

[Caseynotes]

19 minutes ago, dmedin said:

Will need to try 1:2 risk-to-reward with EUR/USD as seem to be missing out on profits when getting it right  

please start a new thread.

[dmedin]

16 minutes ago, Caseynotes said:

please start a new thread.

 

[dmedin]

EUR/USD, hourly time frame, 1:1.5, 100 trades, 50% win rate, 479.4 pips profit but with a significant period of drawdown.  

No point starting a new thread so I'll keep all future results, tweaks and optimizations to myself

 

[Caseynotes]

24 minutes ago, dmedin said:

EUR/USD, hourly time frame, 1:1.5, 100 trades, 50% win rate, 479.4 pips profit but with a significant period of drawdown.  

No point starting a new thread so I'll keep all future results, tweaks and optimizations to myself

 

these figures are meaningless because you have not bothered to state exactly what it is you are testing.

why not start a new thread setting out what it is you are testing and then show the results, - that would actually be useful.

but watch out, there's a troll on this forum who likes to wade in and highjack threads and steer them towards their own ends, ... oh wait 👀

[dmedin]

3 minutes ago, Caseynotes said:

but watch out, there's a troll on this forum who likes to wade in and highjack threads and steer them towards their own ends, ... oh wait 👀

 

Have you been trying to get me banned, Tom? 🤔

[Caseynotes]

please start a new thread.

[Bopperz]

Hey guys,

When back testing a new "idea", what are you doing about the recent market moves since March?

I backtest over multiple years of data, but use a validation/holdout dataset of the final 10% of data points to confirm the idea actually works. However, the recent market moves are starting to make up a larger and larger amount of this validation set.

This means I can create a great model, that generates good PnL over several years, but I should reject it, as it performs poorly due to Covid.

Any questions, let me know!

J

[dmedin]

2 hours ago, Bopperz said:

Hey guys,

When back testing a new "idea", what are you doing about the recent market moves since March?

I backtest over multiple years of data, but use a validation/holdout dataset of the final 10% of data points to confirm the idea actually works. However, the recent market moves are starting to make up a larger and larger amount of this validation set.

This means I can create a great model, that generates good PnL over several years, but I should reject it, as it performs poorly due to Covid.

Any questions, let me know!

J

 

What's a 'validation set'?

Test it 100 times and find a win rate is all I do.

Obviously that is not 'sophisticated' enough  

[Bopperz]

1 minute ago, dmedin said:

 

What's a 'validation set'?

Test it 100 times and find a win rate is all I do.

Obviously that is not 'sophisticated' enough  

Its a bit of data that I never look at, so I cant cheat. The problem is, if you study a price chart long enough you will find something that works. To get around this problem, I test the theory on this unseen data to see if it still performs.

Imagine you were using a SMA cross over strategy and trying out different averaging periods. How would you know that 12 day vs 50 day was the best setup? If you used all your data, you would never know if you had just overfitted to your dataset. Your model may just work really well on past data and not generalise to new data.

[dmedin]

12 minutes ago, Bopperz said:

Its a bit of data that I never look at, so I cant cheat. The problem is, if you study a price chart long enough you will find something that works. To get around this problem, I test the theory on this unseen data to see if it still performs.

Imagine you were using a SMA cross over strategy and trying out different averaging periods. How would you know that 12 day vs 50 day was the best setup? If you used all your data, you would never know if you had just overfitted to your dataset. Your model may just work really well on past data and not generalise to new data.

That's a great idea!

[dmedin]

@Kodiak

In a previous post you said that when testing a system, you should 'exclude' the coronavirus event.

I just wanted to suggest that perhaps you should focus on building a trend-following system.  Such a system will give you the biggest profits when the market is strongly trending.  Hence, you don't need to 'exclude' any 'events'.  You just need to trade your signals.

That's all 🙂

 

[dmedin]

So I really screwed up lately by not following my plan.  Sheer impatience, impulsiveness and stupidity.  I suppose I am not psychologically stable enough to trade patiently and consistently.

I made a new resolution to read over my entire trading plan first thing in the morning before I put any trades on.  Hopefully that will help a bit.

[Bopperz]

17 hours ago, dmedin said:

So I really screwed up lately by not following my plan.  Sheer impatience, impulsiveness and stupidity.  I suppose I am not psychologically stable enough to trade patiently and consistently.

I made a new resolution to read over my entire trading plan first thing in the morning before I put any trades on.  Hopefully that will help a bit.

Can automation help? If the trading plan is solid, and can be expressed in code. It can remove some of the mental problems.

[Caseynotes]

4 minutes ago, Bopperz said:

Can automation help? If the trading plan is solid, and can be expressed in code. It can remove some of the mental problems.

yes automation can help but should not be necessary, the psych issues are caused by a lack of confidence in your system but proper testing will overcome that. If you have proved to yourself that so long as you stick to your rules and the system works on backtesting, forward testing on demo and testing on a live account at minimum position size then you will have the confidence. And though you won't win every trade you'll know you'll win a high enough percentage to be profitable. 

Bots are helpful and can cover many markets 24 hrs but you still need to prove the system actually works in the first place.

 

[dmedin]

16 minutes ago, Caseynotes said:

forward testing on demo and testing on a live account at minimum position size then you will have the confidence

 

I haven't tried 'backesting' trading with pivots.

The basic system that will give a 50% (slightly better or slightly worse) rate could be fully automated I think.  

Buy:

Price > 200 EMA

MACD crosses above signal line from below 0

PSAR is below price

Stop loss: x number of points below PSAR

Take profit: 1.5 times the size of the stop loss

Sell:

Opposite of above

 

But this can be tweaked and modified, because on its own it's very unsatisfactory (it will leave you out of most long-running trends).  I suppose you could have the basic system automated and delve in to do discretionary trading as well.  

I'm sure I could automate and backtest that very basic system using the Pine editor on TradingView 🤓

Edited September 9, 2020 by dmedin

[Caseynotes]

1 minute ago, dmedin said:

 

I haven't tried 'backesting' trading with pivots.

The basic system that will give a 50% (slightly better or slightly worse) rate could be fully automated I think.  

Buy:

Price > 200 EMA

MACD crosses above signal line from below 0

PSAR is below price

Stop loss: x number of points below PSAR

Take profit: 1.5 times the size of the stop loss

Sell:

Opposite of above

 

But this can be tweaked and modified, because on its own it's very unsatisfactory (it will leave you out of most long-running trends).  I suppose you could have the basic system automated and delve in to do discretionary trading as well.  

I'm sure I could automate and backtest that very basic system using the Pine editor on TradingView 🤓

the point of pivots is that they are recognisable boundaries to everyone so you don't want to open a trade going into them but rather going away from them and then use the next one as a target. If you already have an open trade heading towards one be prepared to bail if there is to be a reversal but hope for a charge on through instead.

The system needs to have a precise trigger for entry and then one or two filter indicators who's job it is to keep you out of more bad trades than good ones, it's annoying when they keep you out of a good trade but the important point is the probabilities of outcomes over x number of trades not the result of any one trade.

 

[Caseynotes]

Very good mentoring video from the prop trading firm Axia Futures goes through a trade planning the entry, stop and target conditions on the chart then follows price on the ladder and after the trigger covers in-trade management.

(20 min)

'Why Is Day Trading Futures Not Simple'