Is the concept of momentum and trend trading flawed?

Dr. Toad

Active member
Messages
169
Likes
33
Recently I have been trying to come up with some new systems. When I do this, I often need to quantify concepts. My general thinking of stocks and how price behaves has been for a while now that: A stock that is going up is more likely to continue going up than change directions. The simple reason behind this being that the stock is "trending" and "has momentum". At face value, at least to me, this makes sense. I recently decided to put it to the test though to see if anything can really be gained from trading this way.

I tested this by doing the following:
If a stock has had "x" up days (as defined by an intraday close - open gain) in a row, the stock would be bought the next day at open and sold at close.

I varied "x" from 1 to 10. Obviously in the lower range, I don't think anyone would conclude a stock has momentum or was trending. In the upper range however, this would be the conclusion that I would make. So, my expectation was that in the lower range of "x" values, I would have about a 50/50 win/loss rate, but I figured that if the concepts of momentum and trend truly offer any edge towards trading, I would see at least a marginally higher win percentage with higher "x" values.

I tested this over a 2 year lookback period (7/13/15 to 7/13/17) on 1,430 symbols (~20% of tradeable symbols I have access to). The results are as follows:

Momentum.png

The first column (up days in a row) represents "x". The margin of error based on a 99% probability was calculated as well and is in the last column. From this though it seems pretty clear to me that there is essentially no meaningful edge to be gained from trading with the "trend" or trading when "momentum" occurs. This result distribution would be in line with what would be expected from a pure unbiased coin toss sample.

Anyone have any thoughts?
 
Recently I have been trying to come up with some new systems. When I do this, I often need to quantify concepts. My general thinking of stocks and how price behaves has been for a while now that: A stock that is going up is more likely to continue going up than change directions. The simple reason behind this being that the stock is "trending" and "has momentum". At face value, at least to me, this makes sense. I recently decided to put it to the test though to see if anything can really be gained from trading this way.

I tested this by doing the following:
If a stock has had "x" up days (as defined by an intraday close - open gain) in a row, the stock would be bought the next day at open and sold at close.

I varied "x" from 1 to 10. Obviously in the lower range, I don't think anyone would conclude a stock has momentum or was trending. In the upper range however, this would be the conclusion that I would make. So, my expectation was that in the lower range of "x" values, I would have about a 50/50 win/loss rate, but I figured that if the concepts of momentum and trend truly offer any edge towards trading, I would see at least a marginally higher win percentage with higher "x" values.

I tested this over a 2 year lookback period (7/13/15 to 7/13/17) on 1,430 symbols (~20% of tradeable symbols I have access to). The results are as follows:

View attachment 243058

The first column (up days in a row) represents "x". The margin of error based on a 99% probability was calculated as well and is in the last column. From this though it seems pretty clear to me that there is essentially no meaningful edge to be gained from trading with the "trend" or trading when "momentum" occurs. This result distribution would be in line with what would be expected from a pure unbiased coin toss sample.

Anyone have any thoughts?

There maybe more of an edge if you added the parameter of a breakout past previous highs (13, 26 or 52 wk)
 
Recently I have been trying to come up with some new systems. When I do this, I often need to quantify concepts. My general thinking of stocks and how price behaves has been for a while now that: A stock that is going up is more likely to continue going up than change directions. The simple reason behind this being that the stock is "trending" and "has momentum". At face value, at least to me, this makes sense. I recently decided to put it to the test though to see if anything can really be gained from trading this way.

I tested this by doing the following:
If a stock has had "x" up days (as defined by an intraday close - open gain) in a row, the stock would be bought the next day at open and sold at close.

I varied "x" from 1 to 10. Obviously in the lower range, I don't think anyone would conclude a stock has momentum or was trending. In the upper range however, this would be the conclusion that I would make. So, my expectation was that in the lower range of "x" values, I would have about a 50/50 win/loss rate, but I figured that if the concepts of momentum and trend truly offer any edge towards trading, I would see at least a marginally higher win percentage with higher "x" values.

I tested this over a 2 year lookback period (7/13/15 to 7/13/17) on 1,430 symbols (~20% of tradeable symbols I have access to). The results are as follows:

View attachment 243058

The first column (up days in a row) represents "x". The margin of error based on a 99% probability was calculated as well and is in the last column. From this though it seems pretty clear to me that there is essentially no meaningful edge to be gained from trading with the "trend" or trading when "momentum" occurs. This result distribution would be in line with what would be expected from a pure unbiased coin toss sample.

Anyone have any thoughts?

to add to cbrads comment, i have backtested systems of a break of 52week highs with 70% success rates across the entire FTSE going back 15 years. Trend trading does work, with a significant edge
 
Also if you could define a stop loss level in terms of recent price action, eg. Previous closes, lows, medians, highs etc. then the successful trade's gain could be evaluated in R:R terms and a 50% win rate could still make you money
 
52% took us out of Europe so if it doesn't represent an edge we're in trouble :LOL:

Price doesn't often go straight up without any pullback which is perhaps indicated by the steep fall off in the number of trades as the number of consecutive ups increases. To test the trend it seems to me that you have to cater for temporary pullbacks in some fashion. Quite how I leave to you to wrestle with :)
 
If I am understanding your post correctly, in part at least, you are dealing with the concept of probability. Prior to when you start counting your number of 'up days' you are not paying any attention to what has happened prior to day 1 in terms of determining whether the stock price is in any kind of trend.

Given that on any given day a stock price can go up, stay the same, or go down. With equal weighting of probability, the chance a stock will go up in price is 33%. So if you have calculated that by using your method you can achieve 50%+ wins, then that is pretty good?

As you are aware Im sure, Trend and Momentum do not necessarily occur together. Trend being the direction, Momentum being the speed, while importantly the other part of the equation, in my opinion, is Gravity, hence the likes of consolidation periods, and pullbacks. I think the idea you have is one where the price keeps going up the longer its been going up, whereas if you think of the stock price as a bouncing ball (its the best I can come up with at this time of night...) If you bounce the ball away from you hard onto the floor, it rises up, quickly at first, but then tapers off as gravity takes a hold, before hitting the ground and taking off again in an upward direction etc....The longer its up in the air the less likely it is for it to continue being up in the air without coming down.

A coin as we know has no memory, so statistically it can be flipped for years constantly and each outcome is totally independent of any prior or subsequent outcomes. Something that was touched on in one of the threads earlier this week however, is conditional probablility. I think this may well come into play here because for each additional day the price goes up, the likelyhood of it going up again the following day decreases, as I have just suggested above. This can Im sure be demonstrated statistically, but also consider that unlike the coin flip, sentiment, psychology and a whole load of other factors come in to play with stock prices.

Looking at your thoughts again in your 3rd paragraph. I can see where you are coming from, that each additional day the price goes up is going to confirm a trend. I think your edge, which is seemingly backed up by your spreadsheet is gained by buying a stock that has had the fewer number of updays. Your only consideration is how to screen and choose which ones to buy, which leaves me thinking about stocks for instance that are making new highs? i.e a stock that has gone up in price for one day only, but has also made a new high. Another round of testing maybe...?
 
Last edited:
Wouldn't you consider trend as a longer term thing, i.e. not just the next day/bar. If the trend is up, I wouldn't necessarily think we'll be up on the next day, but maybe up 20 days from now.

Also it's worth thinking that something can be up 50% and down 50% of the time, and still be trending upwards. There is the extent of the move.

Is there overlap in your results? For the case that there were 10 days up and the next was up, does this count as 2 successes for the 9 days in a row system, 3 successes for the 8 days in a row etc.
 
Wouldn't you consider trend as a longer term thing, i.e. not just the next day/bar. If the trend is up, I wouldn't necessarily think we'll be up on the next day, but maybe up 20 days from now.

Also it's worth thinking that something can be up 50% and down 50% of the time, and still be trending upwards. There is the extent of the move.

Is there overlap in your results? For the case that there were 10 days up and the next was up, does this count as 2 successes for the 9 days in a row system, 3 successes for the 8 days in a row etc.

There could be some importance to this very point but I will leave that to you to test.
For instance if the move is growing then the likelyhood is of continuing and conversely if the move is getting smaller then it is likely to reverse.
 
The problem with OP's analysis is that it doesn't take account of the true nature of trends. This has been alluded to by Barjon at #5 and IceMan's #6 bouncing ball analogy. Price trends are a summation of individual price cycles each having phase, amplitude and wavelength and when present take a sinusoidal form. If the ongoing sums of cycle amplitudes are increasing then it is highly likely that an uptrend is present (converse for a downtrend). However, the distance between peaks (wavelength) of the daily cycle which is what I think your analysis was attempting to measure, is unlikely to show the trend.

An eod price chart plot represents all the cycles present (and quite often they may be indistinct, non-existent or exhausted) which is why overall, trendlines are a pretty good guide to trend – easily eyeballed and very low-tech. The difficulty with price trends is that they occur in a sawtooth fashion. You also have to decide what you consider to be a trend e.g. we all know that a one-week uptrend may well be part of a six-month downtrend and this sort of thing can make life very difficult.

Apart from drawing physical trendlines, the only way I have found of determining trends is to measure and sum individual cycles. When a number of cycles are in phase their additive effect is greatest and the future trend can be predicted with reasonable confidence. The caveat is that cycles are not always present and when they are, amplitude wavelength and phase must all be considered carefully.

The best explanation I have found of the nature of trends and cycles within them is in Millard's book "Channels & Cycles" which expands on the work previously done by JM Hurst.
 
Last edited:
I don't recognise what was under the back-test as trend trading.

Then suggest ways in which the OP could test that.

You have to define something before you can test it. For the opening poster - currently - trend is being defined by consecutive up bars. How should he/she define trend?

Well done to the opening poster for actually doing some testing.
 
Then suggest ways in which the OP could test that.

You have to define something before you can test it. For the opening poster - currently - trend is being defined by consecutive up bars. How should he/she define trend?

Well done to the opening poster for actually doing some testing.


I've regularly posted aspects my trend-trading but so have lots of people - no doubt everyone doing it has their own personal approach with many slight differences, so my criteria are no better than any other trend trader's.

I have read a bunch of reports over the years saying that this or that TA method won't work, or even that all TA won't work. But they are usually based on entry, stop and exit criteria that real traders don't use, with no regard to underlying market strength as derived form indices etc., no pyramiding, no scaling in/out, etc. etc. and most crucially no facility to run winners. The classic academics' approach is where the study looks at buying on MA Golden Crosses and demands that the sell only occurs when a Death Cross occurs, and you will be able to find loads of MA studies pdf's that use this ludicrous set-up.

Back-testing is interesting, but only where it is also realistic.
 
I've regularly posted aspects my trend-trading but so have lots of people - no doubt everyone doing it has their own personal approach with many slight differences, so my criteria are no better than any other trend trader's.

I have read a bunch of reports over the years saying that this or that TA method won't work, or even that all TA won't work. But they are usually based on entry, stop and exit criteria that real traders don't use, with no regard to underlying market strength as derived form indices etc., no pyramiding, no scaling in/out, etc. etc. and most crucially no facility to run winners. The classic academics' approach is where the study looks at buying on MA Golden Crosses and demands that the sell only occurs when a Death Cross occurs, and you will be able to find loads of MA studies pdf's that use this ludicrous set-up.

Back-testing is interesting, but only where it is also realistic.

Most approaches using MAs as a primary indicator are flawed because the longer the period of the MA, the longer the period of missing calculation. E.g. a 20 period MA cannot reliably calculate the last 10 points of the data period. This is why Mas should always be plotted in the "centred" fashion i.e. half a period span back, which means that the latest half period span must be estimated – and that's where the accuracy problem lies.
 
Most approaches using MAs as a primary indicator are flawed because the longer the period of the MA, the longer the period of missing calculation. E.g. a 20 period MA cannot reliably calculate the last 10 points of the data period. This is why Mas should always be plotted in the "centred" fashion i.e. half a period span back, which means that the latest half period span must be estimated – and that's where the accuracy problem lies.


Well, you and I and most real traders know that an MA cross-over system is going to give random results, but so many academic papers condemn TA generally using just such a specific set-up.

Of course, these US business schools and university finance faculties which are the source of such a lot of this guff derive no income from private retail traders but such a lot from the investment industry. I guess in some circles, pleasing the client is a good justification for a bad piece of work.
 
welcome to trading ...........nothing works all of the time ......you just have to learn how to read the market and anticipate the moves in advance

N
 
would the op not be better off testing shorter terms within a trend breakout? I mean trend pullbacks /re tracements or buying supports in uptrend .

I am going to unsubscribe.
 
Last edited:
Good discussion so far...not sure what I will be testing next, but to keep the discussion going a bit:


There maybe more of an edge if you added the parameter of a breakout past previous highs (13, 26 or 52 wk)

Your only consideration is how to screen and choose which ones to buy, which leaves me thinking about stocks for instance that are making new highs? i.e a stock that has gone up in price for one day only, but has also made a new high. Another round of testing maybe...?


There is no edge to this either. At least none for the definition of trend I have used. Just finished up some more testing. I didn't really think my results would be any different but it was easy enough to run while I was out. I am including the original set of testing for completeness again in this post. I also ran a baseline with 0 days up (i.e. every stock purchased and sold each day the baseline criteria met).

The other testing I did was the same basic "x" up days, but with an additional requirement of the signal day high must be the 1 month high (~4 week), and the signal day high must be the 3 month high (~13 week).

Momentum.png

Momentum & 1 Month High.png

Momentum & 3 Month High.png

To me the additional baseline parameters do not appear to have any statistically significant impact. As I am essentially looking at a shorter term trend, going out further than 3 months seemed unnecessary. Also, I am thoroughly convinced that there is nothing to be gained by further testing in this manner.


to add to cbrads comment, i have backtested systems of a break of 52week highs with 70% success rates across the entire FTSE going back 15 years. Trend trading does work, with a significant edge

Also if you could define a stop loss level in terms of recent price action, eg. Previous closes, lows, medians, highs etc. then the successful trade's gain could be evaluated in R:R terms and a 50% win rate could still make you money

In order to try and discern if something has an impact on a system, it is important to isolate that thing and test it free of other parameters. I was not so much trying to develop a system with this (although that is always the end goal), but more trying to quantify a concept and check the validity of it. At least to me, this has shown that with a very short term outlook (1 day), trading with the immediate trend has no edge.

I don't think I could ever look at a chart and see the previous 8 days as up days and conclude that the stock is not trending in the short term. Perhaps that it has no momentum, yes. To that point, I probably should not have lumped momentum in with this.

Malaguti: Perhaps there are other parts of your system(s) that are responsible for the positive results? Or perhaps trading with the trend offers some advantage on longer timescales?

Price doesn't often go straight up without any pullback which is perhaps indicated by the steep fall off in the number of trades as the number of consecutive ups increases.

Is there overlap in your results? For the case that there were 10 days up and the next was up, does this count as 2 successes for the 9 days in a row system, 3 successes for the 8 days in a row etc.

Actually, with the original test, the reason that the trades are cut approximately in half each time is quite simply because in the previous iteration, only approximately 50% had up days. So, yes, in the original system if a stock had 11 up days in a row followed by a down day, it would be considered 1 win in the x=10 system, and 10 wins in the x=1 system (and all other variants in between). I wouldn't really consider this overlap since I don't know the next direction in the series (up or down). How would you do it (and why)?


If I am understanding your post correctly, in part at least, you are dealing with the concept of probability. Prior to when you start counting your number of 'up days' you are not paying any attention to what has happened prior to day 1 in terms of determining whether the stock price is in any kind of trend.

Given that on any given day a stock price can go up, stay the same, or go down. With equal weighting of probability, the chance a stock will go up in price is 33%. So if you have calculated that by using your method you can achieve 50%+ wins, then that is pretty good?

The assumption that a stock staying the same price is equally weighted with going up or down is a bad assumption. I have included a baseline test to illustrate this point in this post where each stock was simply bought and sold each day, no filtering. From this there were 6,276 even trades of 423,213 total, or approximately 1.5%. I would say that is a pretty accurate weight that should be given to even days. I attribute the slight upward skew in my results simply to the fact that the market was up overall during my test period. Perhaps that is an incorrect conclusion...makes sense to me though.

As you are aware Im sure, Trend and Momentum do not necessarily occur together. Trend being the direction, Momentum being the speed, while importantly the other part of the equation, in my opinion, is Gravity, hence the likes of consolidation periods, and pullbacks.

Valid point regarding trend and momentum not being the same. The way I have set up my test would be more a test of trend than momentum.

I think the idea you have is one where the price keeps going up the longer its been going up, whereas if you think of the stock price as a bouncing ball (its the best I can come up with at this time of night...) If you bounce the ball away from you hard onto the floor, it rises up, quickly at first, but then tapers off as gravity takes a hold, before hitting the ground and taking off again in an upward direction etc....The longer its up in the air the less likely it is for it to continue being up in the air without coming down.

But if this were true, then my results would indicate that as "x" increases, the % win would drop notably below 50%. This is not the case.

A coin as we know has no memory, so statistically it can be flipped for years constantly and each outcome is totally independent of any prior or subsequent outcomes. Something that was touched on in one of the threads earlier this week however, is conditional probablility. I think this may well come into play here because for each additional day the price goes up, the likelyhood of it going up again the following day decreases, as I have just suggested above. This can Im sure be demonstrated statistically, but also consider that unlike the coin flip, sentiment, psychology and a whole load of other factors come in to play with stock prices.

This is not correct. Although the probability of having 3 up days in a row would be less than 50% (since there are 8 possible outcomes with 3 up days only being 1 of the possible outcomes), the probability of the 3rd day being up would still be 50% (since there are only 2 possible outcomes).


Wouldn't you consider trend as a longer term thing, i.e. not just the next day/bar. If the trend is up, I wouldn't necessarily think we'll be up on the next day, but maybe up 20 days from now.

Trend can be defined on any timescale. I expected that by looking at the immediate stock trend (with up to 10 days, we would be looking at 2 weeks max), I would be able to gain some meaningful insight into the next day/bar. I think I have fairly thoroughly proved that this is not the case (with my definition of trend).

More to the point: if you wait long enough, you can almost always be "right". I suppose I could do another round of testing looking at this on longer time frames.

Also it's worth thinking that something can be up 50% and down 50% of the time, and still be trending upwards. There is the extent of the move.

There could be some importance to this very point but I will leave that to you to test.
For instance if the move is growing then the likelyhood is of continuing and conversely if the move is getting smaller then it is likely to reverse.


This is a good point. Although I didn't really set my tests up to check for this, from looking at the results of the backtests, I would guess that if I did do a proper test of this I would find that the stocks traveled up in magnitude just about as much as down in magnitude...but I can't really conclusively say that. Perhaps I will do a bit more testing properly factoring in momentum...not very motivated to at the moment though.


The problem with OP's analysis is that it doesn't take account of the true nature of trends. This has been alluded to by Barjon at #5 and IceMan's #6 bouncing ball analogy. Price trends are a summation of individual price cycles each having phase, amplitude and wavelength and when present take a sinusoidal form. If the ongoing sums of cycle amplitudes are increasing then it is highly likely that an uptrend is present (converse for a downtrend). However, the distance between peaks (wavelength) of the daily cycle which is what I think your analysis was attempting to measure, is unlikely to show the trend.

Again, I would say trends can be defined on any timescale, but I would agree that the usefulness of them on shorter timescales in naught as demonstrated by this exercise.


The best explanation I have found of the nature of trends and cycles within them is in Millard's book "Channels & Cycles" which expands on the work previously done by JM Hurst.

I might check that out, thanks for that.
 
Good solid work Toad.
If you would like some more possibilities to test let me know and I will try to think of some.
 
Good solid work Toad.
If you would like some more possibilities to test let me know and I will try to think of some.

I'd agree with that.

When it comes to trends, at the risk of stating the bleeding obvious, the easiest and best way to see if a trend is present is just to eyeball it on the chart – and if you can't see it it's not there. No need for fancy calculations. Unfortunately this doesn't help with the other things you need to know about a trend: how long will it last, when is it then likely to change direction and are there any new ones likely? These things only become apparent on the chart post-event. This is the area where study of cycles starts to become useful – not a panacea and needs to be used with a deal of realism. Nevertheless this is the only way I have found to get some idea, some of the time, with some instruments, of trend duration and change.

TA in all its forms is never an exact science and has to be "interpreted". The academics (scientific) who pooh-pooh TA never seem to understand this or the reality that you can be wrong 40% of the time and still be successful. And of course the academics (arts) who sometimes have a more philosophical approach, always seem to be Lefties who won't touch anything smelling of profit with a barge pole and whose main financial skill is in spending what somebody else has earned. (Rant over).

I think Toad's approach is a good one: my experience is that the more you investigate, experiment and work the spreadsheet to death the more you learn.
 
If you would like some more possibilities to test let me know and I will try to think of some.

I'm always searching for things to test out. I have tested out a bunch of stuff in the past, but somewhat aimlessly. I am trying to do more testing with purpose now and with better documentation so that I can hopefully make some forward progress.

I'm a bit short on time at the moment, but feel free to make some suggestions - anyone can. Feel free to post suggestions in my journal (in the process of fixing ATM) or send me a PM.
 
Top