Skate, thanks for detailing, (1) assume you trade them all on ASX only? not US or other markets? we do have good US EOD data from PremiumD and i believe they will be adding a TSX set soon (Canadian market)
(2) Your systems are day/weekly frequency trading but no automated or script running intraday?
(3) Do any of the systems sell short?
I happened to take in Chat with Traders podcast#205 this week - Michael Katz from Seven Points Capital.
205 · Day Trading—The Questions You Want Answered w/ Michael Katz
Michael Katz, equities trader and managing partner of Seven Points Capital, returns to the podcast. This time, answering your questions on: Strategy, Trade Management, Development & Prop Trading.chatwithtraders.com
In Australia we've been fortunate to have Aaron Fifield (Chat with Traders) and Andrew Swanscott (Better System Trader) putting out a lot of fantastic material. Better System Trader podcast Steve Ward talks about the inevitable "J curve" to profitability most traders endure and ways of reducing the depth and time in drawdown before (hopefully) establishing sustainable profitability. Better System Trader Episode 164
@Roller_1 It's easy to backtest the scenario you have put forward if I had Norgate's Platinum Data - unfortunately, I only have the "Silver Subscription" so my records would be skewed. Also, my Norgate data only goes back for 9 years. (2011)
@Roller_1 fair enough.Aren't you worried that you are curve fitting your new systems to your selected time period though or your preconceived bias for how the market works? It isn't really unfortunate it is just a 'small' investment in your trading business really in my opinion. If you test a weekly system from 2018-now it is only approx 150 bars worth of data. Not enough to get a decent sample size of trades imo. I don't think that if someone developed a daily strategy over 6 months of data it could be called thoroughly tested, isn't it a similar situation? They might be great systems and i hope they are but i just don't understand how you can trust them if/when the market changes again?
Aren't you worried that you are curve fitting your new systems to your selected time period though or your preconceived bias for how the market works? It isn't really unfortunate it is just a 'small' investment in your trading business really in my opinion.
If you test a weekly system from 2018-now it is only approx 150 bars worth of data. Not enough to get a decent sample size of trades imo. I don't think that if someone developed a daily strategy over 6 months of data it could be called thoroughly tested, isn't it a similar situation?
They might be great systems and i hope they are but i just don't understand how you can trust them if/when the market changes again?
Agree and in my view 15y is farrrrrr too much.Facts: (a) markets go up, (b) markets go down, (c) markets go sideways. From those 3 facts, which never change, how much data is actually required? What we are actually talking about is the ability to recognise when one state changes into another. The big market moves:
View attachment 115950View attachment 115951View attachment 115952View attachment 115953
Pretty much every move that you would worry about, building a long only system, is encapsulated in 15yrs worth of market data.
Now I don't trade mechanical systems. I don't use software to backtest. Therefore I could be way off base re. the volume of data required. However, what we are talking about is actually human psychology and human psychology coded. We are moving into AI and potentially machines teaching and executing themselves in the market, but we are not quite there yet. Even if we were there, there still remain only 3 available options.
How does more data improve your probabilities?
jog on
duc
Facts: (a) markets go up, (b) markets go down, (c) markets go sideways. From those 3 facts, which never change, how much data is actually required? What we are actually talking about is the ability to recognise when one state changes into another. The big market moves
Pretty much every move that you would worry about, building a long only system, is encapsulated in 15yrs worth of market data.
Agree and in my view 15y is farrrrrr too much.
I mentioned in my thread i do not backtest on data older than 10y old
Why?
If you are old enough, remember 2005 trading market.
Quant trading? Overseas interests in asx?
Quantitative easing?
None of these existed..or nowhere near as at current level.
Hey Ducati,
1. i agree that markets do what you stated but HOW they do that is obviously very important and like you said in the last 15 years there have been a lot of market cycles, great for seeing how a certain system(s) would react over that period.
2. The data is just a mechanism to validate your views of the market, why not use what is available? Skate is only testing on 2 years of data. i guess it's like seeing one earnings report or whatever fundamental guys readand deciding if it's a good company or not. Like Nick Radge say do you want to make money or save money.
Especially in 2019/20 you could probably put a tight stop on any weekly system and get good results (avoid the crash, ride the boom) but in 2015-18 in might have a 40% drawdown from whipsawing in and out of positions.
3. i agree with frogs post below about too much data too, i don't know what the experts say but like from said from 2000-2010 the market was a very different place. But ill still test a system from 07-2010 to see how it goes over an extended crash.
The data is just a mechanism to validate your views of the market, why not use what is available? Skate is only testing on 2 years of data. i guess it's like seeing one earnings report or whatever fundamental guys readand deciding if it's a good company or not.
Is more data better than less? This is really the question
Mr Roller, I have highlighted/underlined your points that I think merit further discussion.
1. First off, HOW markets do what I stated, is exactly how they do it: (a) they go up, (b) they go down, (c) they go sideways. You introduce the market cycle. With which I agree. However, they also only: (a) they go up, (b) they go down, (c) they go sideways. My point being whatever the market environment, they can only do (a), (b) or (c).
2. Is more data better than less? This is really the question. If it is (and I don't work in this milieu) why is it?
3. Different from what and different how? If we are talking about data points revealing price/date and not valuations, fundamentals, news stories etc. (and that should all already be reflected in the price if we believe markets are efficient over time) surely we only have data that goes: (a) they go up, (b) they go down, (c) they go sideways.
jog on
duc
@Roller_1, all your comments are valid when it comes to Amibroker backtesting. The market is completely chaotic & unpredictable due to the infinite amount of information pouring into the markets second-by-second, let alone "year-after-year-after-year".
I'm creating strategies (for my own personal use) with less backtest data & (IMHO) using less data doesn't invalidate my research or backtest results. By "paper trading" those strategies over a period of time will determine if the strategy meets the benchmark to go live. Recent comments about the data required for meaningful analysis varies, such as those you have posted & those posted by @ducati916, @investtrader, @qldfrog & @Warr87 "over the time" are all valid points "in my opinion". The way you handle your own analysis "gives you the confidence" to take a strategy to the next level.
The 'Dump it here' thread is about ideas
I corral my ideas & comments in this thread, ideas that I've found helpful in my trading experience. I'm constantly promoting ideas & sometimes they create a discussion that stimulates us all. Thinking about what is posted is a "valuable part" of this thread & it's a perfect platform for others to express an alternative view or offer a helpful hint. All comments are welcomed, whether the comments are related to system or discretionary trading.
"Skate is only testing on 2 years of data"
Any period of backtesting I post means "Jack" to me & it should mean "Jack" to others as I've previously stated to "ad-nauseam". Price history without "Delisted Equities" & "Historical Index Constituents" causes the inaccuracies in the backtesting results. The results I post should not be relied upon.
We should be all multi-millionaires
Backtesting at times displays outstanding results that "can't" be reproduced in real trading for various reasons. Run a backtest of any good strategy over the last 10 years & we are all multi-millionaires - but in reality, we are far from it.
The 'Dump it here' thread
The 'Dump it here' thread was originally intended to help beginners find their feet, start them off on the right foot (as to say) & is now followed by a few seasoned traders.
Accurate or inaccurate backtesting results
There are times when trading can go south for no good reason that will be unavoidable no matter how good our backtesting results are. Trading has no regards as to how smart we are, or how hard we work, we will regularly be hit by something unforeseen & the next big dip might be just around the corner (who knows).
Trading is Bat-**** scary for beginners & seasoned traders alike
All I'm saying is - "we are not immune to losses" just because we have the "comfort" of a great backtest.
Skate.
How far back you should backtest ??
Here's is an excel snipit of 10 systems backtested 20 years back to Jan 2000.
View attachment 115965
If you look at the bottom row, clearly the best systems over the full 20 years are WTT_AU, MR_US BB_MR.
However, looking closer at the second bottom row ( last 10 years ) MR_US has almost been the worst performer even though it was the best performer from 2000-2010. Clearly it's not working anymore and it's out of the trading plan. MR_AU also is/ was a poor performer. That's not to say that Mean Reversion systems are dead, the last column ( BB MR) is a mean reversion system that i still trade and works quite well in both ASX 300 and SP 500/1500 ( not shown ).
The top row ( weightings ) shows the weights for the systems that i currently value ( ie, better recent performance ). They're a balance between growth and diversity ( non correlation). The last 2 columns on the right in green are (ave) the Average of the systems, (weight) The return of the systems with weighting applied.
The most important aspect of this picture is that NO system is a clear winner. They almost all have negative years, some have stellar years, some are more volatile, some work better in negative markets.
The chart below shows JUST how much of a mess the market is. All of the systems are profitable, but you would be hard pressed to be comfortable with any single one of them. The black dashed line is the average and green weighted.
View attachment 115969
To complete the picture here's another sheet showing some basic monthly metrics and how combining them is an effective strategy.
View attachment 115967
By averaging the systems the losing months drops considerably, volatility is manageable, monthly ave growth is good.
My point is that a lot of insights can be gained by looking at a longer picture and how strategies can fit together into a model. Correlation is not something that can be measured in days, weeks or months but more in years over a broad range of conditions.
As for system development, it is important to pick a time frame, a universe and a goal. 6 months is ample time as markets generally change within 6 months. What is important is that if your system works well for specific conditions, then you need to know when those conditions are met or not and switch that system on or off.
I don't want to discredit your ideas or anything i'm sure you are more successful in trading and life than me, i just feel things could be done better in regards to showing beginners best practices.
Helping Others
You might want to dump stuff here to help others
We use cookies and similar technologies for the following purposes:
Do you accept cookies and these technologies?
We use cookies and similar technologies for the following purposes:
Do you accept cookies and these technologies?