Dump it Here

[Skate]

I'm personally either too old or too lazy to implement something similar

@barney, when we fail to listen, we forgo the right to learn.

Every poster has something to contribute
Barney, I've read most of your posts & at times you tend to sell yourself a bit short.

I'm posting information that worked for me
@ducati916, @peter2, @qldfrog, @Newt, @tech/a, @Warr87, @Joe90, @frugal.rock, @othmana86, @bigdog, @lindsayf, @debtfree, @CNHTractor, @Saqeeb, @martyjames, @Cam019, @rnr, @Rsthree, @MovingAverage @Lone Wolf, & @Beaches all have their own "particular style & trading experience" posting in a manner that they feel will be helpful to others. (sorry if I've missed a contributor to this thread, I've gone from memory)

Sharing is caring
Without realising it, we are all in this game together just playing different positions on the same team. (even if at times it doesn't feel that way)

Skate.

 

[Wilham]

@barney, when we fail to listen, we forgo the right to learn.

Every poster has something to contribute
Barney, I've read most of your posts & at times you tend to sell yourself a bit short.

I'm posting information that worked for me
@ducati916, @peter2, @qldfrog, @Newt, @tech/a, @Warr87, @Joe90, @frugal.rock, @othmana86, @bigdog, @lindsayf, @debtfree, @CNHTractor, @Saqeeb, @martyjames, @Cam019, @rnr, @Rsthree, @MovingAverage @Lone Wolf, & @Beaches all have their own "particular style & trading experience" posting in a manner that they feel will be helpful to others. (sorry if I've missed a contributor to this thread, I've gone from memory)

Sharing is caring
Without realising it, we are all in this game together just playing different positions on the same team. (even if at times it doesn't feel that way)

Skate.

Great conversation going on here. I'm sitting on the sidelines taking it all in as I'm still several steps behind you all in experience here.

 

[Skate]

This is a very interesting topic and one that I'm sure has the potential to generate another flurry of posts. "Returns from trading started to become the scorecard". Yes, returns is an important measure, but for me it is a secondary important measure and doesn't top my scorecard measure. For me personally my priority is to chase a nice tight standard deviation in returns. Why, it delivers a better level of predictability and lessens surprises. Yes, a tight standard deviation generally comes at the cost of a higher return, but I also don't like the wild ride that comes with surprises.

Trading means something different to everyone
Back when I was in system development "tight standard deviation in returns" (the scattergun chart of returns) was important as a tight range eliminated outliers (to some degree) but the reference @ducati916 made about shifting market forces (my wording) making an extreme amount of sense. In saying this - the very points that you have made are just as sound. (With each point of view put forward "the poster" always slants the presentation in their favour)

Trading signals
When it's working, (in relation to trading) I tend to keep working it. I try to post (ideas) to stimulate the thinking of others. I'm not saying "Do as I do" but for others to "Think about what they are doing" - that's all. (my views are from my perspective)

Skate.

 

[Skate]

Great conversation going on here

@Wilham, I knew I'd miss someone from memory (thank you for your contribution to this thread)

Skate.

 

[Skate]

I had planned to make a series of posts
Thinking more deeply about the recent flurry of posts, I've come to the realisation there is no point in detailing an answer about an activity with so many variables (the very point @ducati916 alluded to).

In defence of "system trading"
The "Dump it here" thread is full of practical examples of buying into an uptrend. Trading this way "sometimes it works". In my case, having a 40% strike rate works for me which is enough to make trading a profitable pastime.

Mechanical trading
Trading systematically is a series of scenarios: "if this happens - do this" but only if it meets or exceeds certain parameters. Even the best idea with the wrong filters & parameter settings will struggle to be profitable.

Exiting
Knowing when to get off the position (exiting) is the "big one" to sort out - as far as I'm concerned.

Having a great exit strategy at times is not enough
This is the very point raised by @ducati916 but exiting the way I do is the best exit I have. In my defence, the exit signals are "consistent" (as @peter2 commented - "tongue in cheek"). With trading, there is always a trade-off - "you win some, you lose some". When you accept this, you start to understand trading.

Trading is a numbers game
Casinos are also a numbers game but with trading the more you play the better you become.

Skate.

 

[MovingAverage]

In defence of "system trading"
The "Dump it here" thread is full of practical examples of buying into an uptrend. Trading this way "sometimes it works". In my case, having a 40% strike rate works for me which is enough to make trading a profitable pastime.

This is a good point. And reminds me that some systems traders often develop and trade a single system across many different market conditions--up, down and sideways. Problem is getting a single system to deal efficiently with all those market conditions is a bit of a holy grail. Kind of like a committee setting out to design a race horse, but in the end it finishes up with a camel. Too many compromises and a system that can be somewhat average. For me I've had much greater results trading very different and separate systems for different market conditions.

 

[Wilham]

This is a good point. And reminds me that some systems traders often develop and trade a single system across many different market conditions--up, down and sideways. Problem is getting a single system to deal efficiently with all those market conditions is a bit of a holy grail. Kind of like a committee setting out to design a race horse, but in the end it finishes up with a camel. Too many compromises and a system that can be somewhat average. For me I've had much greater results trading very different and separate systems for different market conditions.

If you can, would you mind elaborating on this? My thought is that the common problem of when and for how long is the market going to trend isn't solved by this. I'd be interesting to understand how you turn on and off your systems accordingly? Alternatively are your systems across different universes?

 

[MovingAverage]

If you can, would you mind elaborating on this? My thought is that the common problem of when and for how long is the market going to trend isn't solved by this. I'd be interesting to understand how you turn on and off your systems accordingly? Alternatively are your systems across different universes?

I use an overarching TSI indicator (Trend SI and not the True SI indicator) to determine the general trend of the market. If you're an AB user you can find the AFL for it in their library. I'm over simplifying things for the sake of keeping it simple, but this indicator serves me well in switching my systems in an out. I "turn on" my breakout system when the TSI indicates a broader upward direction and I "turn on" my swing system when the TSI suggests the market is moving sideways. I don't short so my systems tend to remain inactive when TSI suggests a broader downtrend. I guess there are any number trend strength indicators you could use for the broader market, but I've been using TSI for several years with reasonable results. In relation to your universe question, I should add by break out system trades only XAO with turnover filter applied and my swing system trades all listed equities with turnover filter applied

 

[Skate]

I use an overarching TSI indicator (Trend SI and not the True SI indicator) to determine the general trend of the market.

If you can, would you mind elaborating on this? My thought is that the common problem of when and for how long is the market going to trend isn't solved by this.

Let's talk about a momentum indicator
@MovingAverage posted that he uses the TSI indicator (an oscillates that moves between 100 & -100) to gauge the strength of a trend. I've recently posted that there is a multitude of indicators you can use for a trend entry into a position & the TSI indicator certainly fits the bill. Admittedly the TRI Indicator does a half-decent job as a momentum indicator - but there are many applications for this indicator.

The TSI can provide an early indication of a trend
The TSI can also provide an early indication of whether the prevailing trend will continue or reverse but I want to discuss one way that I use the indicator as the basis of a "trading system".

So how can we trade the TSI momentum indicator?
As the TSI indicator is an oscillator, oscillating between (100) & (-100) we can use the zero-line as our base. As the TSI oscillator moves from a negative number to a positive number (basically meaning crossing from below zero to above) we can use this as the activation of our buy condition. Our aim is to get into a trend & the TRS indicator moving above the zero lines does this. Add parameters & filters to your buy condition & then you have a trend strategy.

Summary
When the TSI indicator crosses above the zero-line this is a sign that the individual position has turned "bullish". By the way, trading this way eliminates the use of an Index Filter. By contrast, when the TSI crosses below the zero-line, this is a sign the price action has turned "bearish".

Just keep in mind
If you change the lookback period, length, or smoothing periods, the indicator will respond to price action accordingly. Using the TSI as an exit strategy is not advisable (as far as I'm concerned) as at times the indicator is slow to react - even though this is a "Low Lag Indicator"

Forget the backtest results (They aren't that important)
The "backtest results" are posted to indicate that "any entry into a trend" can be profitable.



Skate.

 

[Lone Wolf]

I use an overarching TSI indicator (Trend SI and not the True SI indicator) to determine the general trend of the market. If you're an AB user you can find the AFL for it in their library.

Is it actually called the Trend Strength Indicator (TSI)? I've been searching google and the Amibroker library for 45 minutes and can't find it. There's one post on the Amibroker forum, but that's not an oscillator like Skate suggests.

 

[Wilham]

If we can set aside the MC and statistic significance for a moment (if that is possible given the below? ) I would really be interested in this groups discussion on how to test a system. The majority of research I've done states you need to optimize with IS data, test it over OOS data then perform WF testing of the entire system (or how I've understood it). When designing a system, if I use a simplified example of BBO I'm playing with at the moment how do you go about this in practice?

Say I have 4 variables I'm trying to optimize:
Var 1: Period of the moving average I use for the index filter
Var 2: BB MA period
Var 3: BB Deviations
Var 4: Trailing stop 20%

Using either the full data set I have for ASX, or going by the 2006-present discussion. I can optimise my variables and test them out of sample without overlapping. Then have a period of WF testing of the full system. The first example I can test my potentially less 'relevant' variables over the more recent data that includes a wide range of market types. The latter example I could use the sample method and have less WF timeframe available, or just paper trade it live.



I've been using the top method. My rationale being following a trend should be similar through time in terms of characterists. If I can select robust variables that turned an acceptable CAR/MDD from 1992-2005 and then test this over 2006-2020 with accpetable returns I can be reasonably comfortable that it would continue to work going forward.

Couple questions as I think I've missed some things in my understanding:
1. Without inherently 'fitting' my optimisations, how do I go back and tweak my system or variables in a design, built, test loop? Is it by finding robust variables that are profitable around my 'best' value. e.g. Var 2 of 100 days, where 80-120 all return a CAR/MDD result that is varying levels of favourable?
2. Is the above even a good way to optimize a system, appreciating people have person preferences etc.
3. Can your IS time periods overlap, e.g. 1992-1994, then 1993-1995 and so on..

The point of this isn't which is better, but I haven't seen much discussion here or on the internet on HOW people with successful systems do this.. usually people in my situation who are far from experts. I feel like a primary school student with the professors but I'm hoping some other students are sitting behind me listening in as well.

 

[MovingAverage]

Is it actually called the Trend Strength Indicator (TSI)? I've been searching google and the Amibroker library for 45 minutes and can't find it. There's one post on the Amibroker forum, but that's not an oscillator like Skate suggests.

Trend. I’ll post up the AB link (think I should be able to do that) a bit later today

 

[MovingAverage]

If we can set aside the MC and statistic significance for a moment (if that is possible given the below? ) I would really be interested in this groups discussion on how to test a system. The majority of research I've done states you need to optimize with IS data, test it over OOS data then perform WF testing of the entire system (or how I've understood it). When designing a system, if I use a simplified example of BBO I'm playing with at the moment how do you go about this in practice?

Say I have 4 variables I'm trying to optimize:
Var 1: Period of the moving average I use for the index filter
Var 2: BB MA period
Var 3: BB Deviations
Var 4: Trailing stop 20%

Using either the full data set I have for ASX, or going by the 2006-present discussion. I can optimise my variables and test them out of sample without overlapping. Then have a period of WF testing of the full system. The first example I can test my potentially less 'relevant' variables over the more recent data that includes a wide range of market types. The latter example I could use the sample method and have less WF timeframe available, or just paper trade it live.

View attachment 121545

I've been using the top method. My rationale being following a trend should be similar through time in terms of characterists. If I can select robust variables that turned an acceptable CAR/MDD from 1992-2005 and then test this over 2006-2020 with accpetable returns I can be reasonably comfortable that it would continue to work going forward.

Couple questions as I think I've missed some things in my understanding:
1. Without inherently 'fitting' my optimisations, how do I go back and tweak my system or variables in a design, built, test loop? Is it by finding robust variables that are profitable around my 'best' value. e.g. Var 2 of 100 days, where 80-120 all return a CAR/MDD result that is varying levels of favourable?
2. Is the above even a good way to optimize a system, appreciating people have person preferences etc.
3. Can your IS time periods overlap, e.g. 1992-1994, then 1993-1995 and so on..

The point of this isn't which is better, but I haven't seen much discussion here or on the internet on HOW people with successful systems do this.. usually people in my situation who are far from experts. I feel like a primary school student with the professors but I'm hoping some other students are sitting behind me listening in as well.

Here is my 2 cents worth...testing is a two step process: 1) tune (optimize) on your in sample data; 2) test on out of sample data. Personally, walk forward I wouldn't worry about because in AB it continually tunes your parameters on in sample and uses that on the next out of sample block. I don't understand this process because their is generally a low correlation. Make sure your out of sample testing has roughly the same amount of trades and your system generated on in sample. Your out of sample testing does not need to be consecutive years following your in sample--e.g., if you in sampled over 1992 to 1995 your out of sample does not need to be the consecutive blocks of 1996 - 1998, 1999 - 2001 etc. What is important is that you just use any data outside. In this regard, I will typically randomise the date range of the out of sample data. For example, I might randomly select the first out of sample range to be July 2001 to Feb 2004, Oct 2004 to March 2007 and so on. Also, if you have access to US market data run your system (do not tune) over that data to see how it performs. This is another form of out of sample data. The main focus of out of sample testing is to ensure it continues to return a positive EV.

In relation to your question 1, I also start on out of sample using the MC mean values for my out of sample data. As for your question 3, I stay away from using any overlap because I like there to be a low correlation between in and out data

 

[MovingAverage]

Is it actually called the Trend Strength Indicator (TSI)? I've been searching google and the Amibroker library for 45 minutes and can't find it. There's one post on the Amibroker forum, but that's not an oscillator like Skate suggests.

Here is a link to the original article: https://engineeringreturns.wordpress.com/tsi/

I've made a few tweaks to my implementation, but it is fundamentally the same.

 

[MovingAverage]

If we can set aside the MC and statistic significance for a moment (if that is possible given the below? ) I would really be interested in this groups discussion on how to test a system. The majority of research I've done states you need to optimize with IS data, test it over OOS data then perform WF testing of the entire system (or how I've understood it). When designing a system, if I use a simplified example of BBO I'm playing with at the moment how do you go about this in practice?

Say I have 4 variables I'm trying to optimize:
Var 1: Period of the moving average I use for the index filter
Var 2: BB MA period
Var 3: BB Deviations
Var 4: Trailing stop 20%

Using either the full data set I have for ASX, or going by the 2006-present discussion. I can optimise my variables and test them out of sample without overlapping. Then have a period of WF testing of the full system. The first example I can test my potentially less 'relevant' variables over the more recent data that includes a wide range of market types. The latter example I could use the sample method and have less WF timeframe available, or just paper trade it live.

View attachment 121545

I've been using the top method. My rationale being following a trend should be similar through time in terms of characterists. If I can select robust variables that turned an acceptable CAR/MDD from 1992-2005 and then test this over 2006-2020 with accpetable returns I can be reasonably comfortable that it would continue to work going forward.

Couple questions as I think I've missed some things in my understanding:
1. Without inherently 'fitting' my optimisations, how do I go back and tweak my system or variables in a design, built, test loop? Is it by finding robust variables that are profitable around my 'best' value. e.g. Var 2 of 100 days, where 80-120 all return a CAR/MDD result that is varying levels of favourable?
2. Is the above even a good way to optimize a system, appreciating people have person preferences etc.
3. Can your IS time periods overlap, e.g. 1992-1994, then 1993-1995 and so on..

The point of this isn't which is better, but I haven't seen much discussion here or on the internet on HOW people with successful systems do this.. usually people in my situation who are far from experts. I feel like a primary school student with the professors but I'm hoping some other students are sitting behind me listening in as well.

One other point and sorry if this is turning into a class in statistics. But what I like to do is plot the distribution of system's key performance metrics (e.g., CAGR, DD, W/L ratio etc) from the in sample testing from MC. Then I take the same key performance metrics from a single run on the out of sample data and see where they fall within the distributions. If the single run metrics fall with 2SDs of the distributions it gives me a much better level of confidence on the system's performance on future data. If the single run metrics are outside the 2SD my level of confidence in the system's ability to perform as expected on future data is very low and I go back to the drawing board

 

[qldfrog]

Here is a link to the original article: https://engineeringreturns.wordpress.com/tsi/

I've made a few tweaks to my implementation, but it is fundamentally the same.

dead link for afl code , but I found
function TSI()

{
Ratio = abs(Close - Ref(Close,-10)) / ATR(10) ;
Result = MA(MA(Ratio,10),100);
return Result;
}
hope it helps

 

[MovingAverage]

dead link, but I found
function TSI()

{
Ratio = abs(Close - Ref(Close,-10)) / ATR(10) ;
Result = MA(MA(Ratio,10),100);
return Result;
}
hope it helps

That's pretty much it. There are some interesting discussions on another forum about this indicator and a few interesting suggestions for refinement. You can also add a signal line to the TSI (which I use in conjunction with the TSI), which is the main difference between my version and the TSI as originally proposed

 

[qldfrog]

That's pretty much it. There are some interesting discussions on another forum about this indicator and a few interesting suggestions for refinement. You can also add a signal line to the TSI (which I use in conjunction with the TSI), which is the main difference between my version and the TSI as originally proposed

Sorry MA, not sure I understand what you mean about signal line?are you saying a line of code about signal?, a line(threshold)? Sorry for the stupid question :-(

 

[MovingAverage]

Sorry MA, not sure I understand what you mean about signal line?are you saying a line of code about signal?, a line(threshold)? Sorry for the stupid question :-(

Like the MACD indicator in AB...there is the MACD line and there is a Signal line. I have incorporated a Signal line into my TSI

 

[qldfrog]

Like the MACD indicator in AB...there is the MACD line and there is a Signal line. I have incorporated a Signal line into my TSI

for newbies like me: whar is signal? in MACD case:
The signal line is a 9-day EMA of the MACD line. As a moving average of the indicator, it trails the MACD ....
Basically a smoother view of the indicator to remove noise.Dummies like me need full explanation