A key property of a market

To add comments, please log in or register
Elroch
143
Elroch  

All trading (except when based on external information) relies (consciously or not) on inferences about the statistical properties of the market in the future from statistical properties of the market in the past. It seems useful to look at statistics more general than the results of a particular trading system (which seems by far the most common statistical analysis undertaken by traders).

One key property of the market is whether it is trending. (Of course this is a loose term that can be defined in many different but related way). Traders often talk about taking into account whether a market is trading in making trading decisions. They are mainly interested in whether the market will be trading in the future, over a time scale similar to a trade they might make. But the information they have is about the market's past, say that it has been trending over a certain period.

My question is whether anyone else here has done analysis that determines how well they can predict whether the market will trend over a chosen period of the future, using information from the past? Eg if the market is trending over period A in the past, it is more likely to trend over period B in the future. Note that completely random, untradable markets exhibit apparent trends (probably almost as much as real markets), so this comparison will be the benchmark for prediction.

If it is possible to show that it is possible to predict (to a useful degree) whether the market will trend (as one would hope and expect), it would be useful as a component of trading systems. In fact a lot of trading systems use such a condition already, whether or not it is known how reliable it is.

A key aspect of this concept is that it is not directional. Of course all normal trading is really about predicting directions. However, if one knows a market is trending but not even any information about the direction, it may be possible to make a profit by using an appropriate exit strategy. Most of the time there are reasons to have some belief about the direction of the market as well, which will make things better.

To make this quantitative requires answering some questions. How would you define quantitatively whether a market is trending on a particular scale (or to what degree it is trending)? What methods are suitable for predicting whether a market will trend over a period in the future, according to this chosen definition (and can be shown statistically to be so)?

Anyone else interested in these issues, or this sort of general statistical analysis of markets?

zzuegg
1502
zzuegg  
To make this quantitative requires answering some questions. How would you define quantitatively whether a market is trending on a particular scale (or to what degree it is trending)? What methods are suitable for predicting whether a market will trend over a period in the future, according to this chosen definition (and can be shown statistically to be so)?

To estimate the trading in the near past a regression channel could be used. Depending on the slope of the channel even the degree can be numerized.

Predicting the market, well, should not be that easy. If there is a common known way i think we would not have this discussion. You can assume that the market will start trending in the near future when you are into a ranging phase. (But as current eurusd shows, this ranging period can be extended). From my point of view it all comes down to the quality of your entry/exit signal.

If you belive there is an underlying formula/system/patterns in the market then neural nets or statistics might be the best way to "predict" the market.

Elroch
143
Elroch  

I like the idea of the regression line slope. As the axes have different units, it would be necessary to use something like the ATR as a conversion factor, so that the angle could be defined. However, if you use this definition, you would need to be happy with a V-shaped recent price chart being classified as not trending, as the angle of the regression line would be zero! A slightly more complex version would avoid this. What you can do is to use your idea for many segments of the price chart rather than one. For example, you could find the angle of a 10 bar regression channel that ends at each of the last 50 bars. So you'd have 50 angles indicating how sharply up or sharply down the market had been over 10 bar subsets of the last 50-60 bars. You could find the average of the absolute values of these angles, or something like the root mean square of the sines of these angles. The advantage is that a market that had gone sharply down, then sharply up (V-bottom) would be identified as having been trending rather than sideways. A non-trending market would tend to have a lower average absolute angle than a trending one.

I also wondered whether a lack of trending could be a predictor for trending in the future. But I thought it more likely that a long period of non-trending broken by a short period of trending might be predictive of trending for a while (as in breakout strategies). I also think a long period of trending would be predictive of a short period of trending, on the grounds that conditions tend to persist. But I believe the same applies to non-trending - a market that has been non-trending (maybe ranging) for a long while is probably predictive of non-trending for a short while in the future.

Predicting the behaviour of the market (as opposed to where it will go) is something all successful traders have to do to some extent even if they don't realise it. If you think about it, the prediction "my trades, executed according to my rule, have a positive expectation" is a statistical prediction about market behaviour. It is a prediction that the market has certain properties that are not entirely random (in the sense that Brownian motion is random).

A market that has a tendence to trend according to some definition is essential to make any system work. If the market has a tendency to move in one direction for longer than say Brownian motion, it makes it profitable to let a trade run and take bigger profits, if you are lucky enough to trade in the right direction. And if you have the direction long, it means exiting fairly quickly with a stop avoids bigger losses. Even trading systems that claim not to rely on a trend generally do. Say if you are trading a range and sell near the top, you are essentially relying on the market having a tendency to trend away from the edge rather than just wander in a directionless manner - it's just on a different scale. A market that moved like Brownian motion would be a non-trending market that would wander rather slowly, but would be entirely impossible to trade profitably precisely because of it's lack of any tendency to trend.

I have not the slightest doubt that there are underlying statistical properties of the market that allow profitable trading. This is proven by the statistical results of some traders, which cannot be explained by chance. But these properties undoubtedly change with time. Even "second level" properties such as the tendency of a market to trend in the future based on its behavior in the past may change over time. But clearly there are properties that are sufficiently stable to make it profitable to rely on them.

[In the above post, I am using the phrase "Brownian motion" in the mathematical sense]

zzuegg
1502
zzuegg  
Elroch:

I like the idea of the regression line slope. As the axes have different units, it would be necessary to use something like the ATR as a conversion factor, so that the angle could be defined.

Why not use percentage based charts? This would eliminate the 'lagging' of ATR and the need for a normalization.

However, if you use this definition, you would need to be happy with a V-shaped recent price chart being classified as not trending, as the angle of the regression line would be zero!

No, i would consider a market ranging when a M or W shaped price patter has emerged.

You could find the average of the absolute values of these angles, or something like the root mean square of the sines of these angles. The advantage is that a market that had gone sharply down, then sharply up (V-bottom) would be identified as having been trending rather than sideways. A non-trending market would tend to have a lower average absolute angle than a trending one.

Sure, but this fairly complex calculation does not allow to predict anything. All you get is the market charactersitis for the subsets, and when averaging the total period. (Should not be more information than analyzing a moving average over that period, or even the absolute price change in percent. (Quite easier to calculate)

A market that has a tendence to trend according to some definition is essential to make any system work. If the market has a tendency to move in one direction for longer than say Brownian motion, it makes it profitable to let a trade run and take bigger profits, if you are lucky enough to trade in the right direction. And if you have the direction long, it means exiting fairly quickly with a stop avoids bigger losses. Even trading systems that claim not to rely on a trend generally do. Say if you are trading a range and sell near the top, you are essentially relying on the market having a tendency to trend away from the edge rather than just wander in a directionless manner - it's just on a different scale. A market that moved like Brownian motion would be a non-trending market that would wander rather slowly, but would be entirely impossible to trade profitably precisely because of it's lack of any tendency to trend.

I have to say that i do not understand the deepness of the brownian motion, but i agree that the market is aways trending (if you scale the scale) (maybe beside some minute on off hours)

I have not the slightest doubt that there are underlying statistical properties of the market that allow profitable trading. This is proven by the statistical results of some traders, which cannot be explained by chance. But these properties undoubtedly change with time. Even "second level" properties such as the tendency of a market to trend in the future based on its behavior in the past may change over time. But clearly there are properties that are sufficiently stable to make it profitable to rely on them.

Thats why we are here.

From my trading experience i have to say that stock trading, even indices seem more constant to me. I don't know why but i assume its due less noise, less speed and more transparent informations. I assume there are not that many self made millionairs in forex trading.

I also wondered whether a lack of trending could be a predictor for trending in the future. But I thought it more likely that a long period of non-trending broken by a short period of trending might be predictive of trending for a while (as in breakout strategies). I also think a long period of trending would be predictive of a short period of trending, on the grounds that conditions tend to persist. But I believe the same applies to non-trending - a market that has been non-trending (maybe ranging) for a long while is probably predictive of non-trending for a short while in the future.

I think there are too many faktors like economic news, crisis, earthquakes, nuklears fallouts and political movements so that long term statistics can exploit the market.


My 2 cents

Ubzen
5299
Ubzen  

@Elroch: Here are my Opinions.

1) predict whether the market will trend over a chosen period of the future, using information from the past? Yes, every1 here is doing just that. My most direct approach was using Predictive indicators and those Fail flat on their backs.

2) How would you define quantitatively whether a market is trending on a particular scale (or to what degree it is trending)? You can 100%, all the time based on your own narrow definition of Trending upon historical prices. You cannot on future prices.

3) What methods are suitable for predicting whether a market will trend over a period in the future, according to this chosen definition (and can be shown statistically to be so)? The price itself decides. Use flexible and complex trailing stop-Techniques while allow price to move in your direction to the heavens. Well almost the heavens, use Exit strategies based on your own narrow definition of Over-bought/Over-Sold or Whatever makes you feel less greedy.

Currently, I don't feel there's any Bullet-Proof method of trading. If a trading strategy can lose, the question is not if it'll lose-everything but when. My current statistics focus on analyzing the gains vs losses to get ahead. This takes the focus off looking for Bullet-Proof systems and places the focus on which one of my decent system is better for me to use.

@Zzuegg: We think so much alike, someday we should form a team.

1) To estimate the trading in the near past a regression channel could be used. Depending on the slope of the channel even the degree can be numerized. I fully Agree. I'll attempt to simplify it even more. One could just use a Straight-Horizontal-Line with X-Periods ago (based on your narrow defination). If it slopes then you have a trend and if not (based on ur def again) then you're in a range.

Again: just my Opinions, I've said it once and I'll say it again. 99% of all similar indicators gives off the same conclusions. By similar I mean trend indicators will give off the same info as another trend indicator. A range indicator will give similar information as a different range indicator. Usually the only difference between is adjusting the X-Periods. If you think MACD's slow then try using 5-Periods. If you think Williams% is too fast then try using 1000 periods.

The markets only have 2 states (tho Subjective) and that's Trending or Ranging. Therefore, IMO One only need 2 Indicators a Trending-Ind and a Ranging-Ind. Currently, I only utilize 2 indicators (No Not Macd, tho its my Favorite :)),

A Regression-Like Indicator called Center of Gravity (Modified) for Ranges which gets Fatter or Shrinks as price Trends or when the Range becomes Violent. I don't like ATR (tho it's the most popular by the senior members on this forum), reason I don't like ATR is because it Decreases as the price becomes Skinny. I don't mind this when price is in a Flat-Narrow Tunnel. Problem is ATR decreases when price is Rising or Dropping like hell - If the fall and rise is NOT in a Violent Zig-Zag while Trending. My Regression-Channel does Not have this problem. As far as the V shape, I usually consider that a Range-Definition, No different than if the Regression-Channel was Flat.

A CCI based Indicator created by Zzuegg with-in the code-base called Osc. I liked this indicator the first time I saw it because I wanted a Heikin-Ashi type indicator which provided more measurable information. I believe it's multi-time frame and I use Fibonacci sequences for the periods just for the heck of it. This is my main Trend Indicator, It was modified into my Multi-Currency Indicator.

I'm not saying these are the best Indicators as the major reasons I selected them is because I loved that they made the charts look Pretty and Per my belief any other indicator wouldn't make much difference. IMO the key really comes down to a combination of Techniques- Allot of different techniques. Ultimately, the trader's mindset/world-view and general attitude toward Risk and Money will decide the type of trader he/she'll become. So I don't think there's any real Key which fits all.

Elroch
143
Elroch  

@zzuegg,

Why not use percentage based charts? This would eliminate the 'lagging' of ATR and the need for a normalization.

Percentage based charts don't entirely solve the issue of getting an indication of "trendiness" that is independent of time frame, numbers of bars, etc. A good analogy is the stochastics indicator which successfully defines overbought or oversold independently of all these things. MACD by comparison cannot be used for this purpose, because it is not unitless. Also, note that using the ATR of the last N bars in a calculation to determine whether the market has been trending over the last N bars introduces not the slightest extra lag.

No, i would consider a market ranging when a M or W shaped price patter has emerged.

If we agree that a V-shaped chart is not best classified as sideways, we want a trend indicator to indicate this, so something more complex than a single regression line slope (or a single value of ROC) is needed (as the regression line for a V is flat, and the ROC is also zero). The idea I described of looking at lots of short sections of the data is not difficult to implement - I did something similar using ROC instead of linear regression a while back.


Sure, but this fairly complex calculation does not allow to predict anything. All you get is the market charactersitis for the subsets, and when averaging the total period. (Should not be more information than analyzing a moving average over that period, or even the absolute price change in percent. (Quite easier to calculate)

Thats why we are here.

I firmly disagree that the calculation I described gives the same as a moving average or a ROC. It's simply quite different, and specifically distinguishes different market conditions. For example, ROC of a V is 0, as is the linear regression slope. However, the indicator I mentioned writing earlier used ROC for shorter periods and combined these. By doing so, it can distinguish a V from a flat market. The same would be true of an indicator that combined linear regression on lots of short sections of the V. But the main point of this thread is precisely to discuss how well this sort of analysis can be used to predict the statistical properties of the market over a short time in the future. Proper empirical work and analysis is needed.

I think there are too many faktors like economic news, crisis, earthquakes, nuklears fallouts and political movements so that long term statistics can exploit the market.


Not quite sure what you mean here. I have only been discussing statistics that are short term. I believe such statistics have statistical predictive power, but that this predictive power is not for ever - the market behaviour changes, fast in some ways, slow in others. But the whole point of this topic is about finding out how well it is possible to predict statistical properties of the future market based on statistical properties of the past market. Profitable tading can only be possible if such prediction is possible to some extent.

A key point is that every technical trader is making statistical predictions, whether they are conscious of it or not. A trader has seen the past behaviour of the market, has come to beliefs about how this behaviour influences the likelihood of future behaviour and uses this to make decisions that they hope have positive expectation [For example a type of system has been profitable in the past, so it is believed it will be profitable in the future]. So belief in it being possible to trade profitably implies a belief in the market having exploitable statistical properties. I am just being explicit about this.
To add comments, please log in or register