Jump to content
Perishabull

PositiveDev's trading journey

Recommended Posts

I've been away a few days so to update trading since 21st July;

 

23rd July

E-Mini Dow Jones futures
23LHF_zps84b0bf94.png

1 trade, 1 win

 

24th July

E-Mini Dow Jones futures

24LHF_zps574e5717.png

 

1 trade, 1 win

 

25th July

E-Mini Dow Jones futures

25LHF_zps0cb3cc2f.png

 

1 trade, 1 loss

 

28th July

E-Mini Dow Jones futures

28LHF_zpsbe7f6f9f.png

 

2 trades, 1 win, 1 loss

 

29th July

E-Mini Dow Jones futures

29LHF_zpsc0b70c19.png

1 trade, 1 loss

Share this post


Link to post
Share on other sites

Did US equity markets top on 24th July?

 

US equity market comparison

marketsLHF_zps3c27aeaf.png

 

There is a very clear disconnect between the NASDAQ 100 and the Russell 2000 spanning across July.

 

 

This next chart is S&P500, with Australian dollar, Euro, and AUD/JPY (futures markets);

2LHF_zpsa288ac4b.png[/

 

Very clear divergence between these markets with a top made on the 24th July.

 

The NASDAQ 100 looks like the candidate to short.

Share this post


Link to post
Share on other sites

Unfortunately I've found a flaw in this trading strategy I've been using of late that means it's not wise to continue with it. I need to be more rigorous with my research and testing work in the future. I think some of the principles I've learned are valid and relevant, but the way these are tested, applied and executed need to be far sharper.

Share this post


Link to post
Share on other sites

Hmm.

Can you say a little about the nature of the flaw?

Share this post


Link to post
Share on other sites

I've been doing some really interesting work recently in an effort to define an edge within the market. Through the middle of last year to the start of this year I had been successfully trading a strategy (Strategy I) using an edge that I thought I had developed based on an underlying principle I had formed on some factors that influence short term price movement in stock index futures markets. That strategy broke down and instead of having a majority positive profitable days, it changed to a majority of negative losing days, therefore I had no option but to stop trading it. I never managed to understand why that happened, however based on recent work I think I may have now found the answer.

 

I also had traded a further strategy in recent weeks (Strategy II), which looked highly promising however calculation issues meant that I could take a signal to trade, find it to be a losing trade, then recalculate the data at day end to find myself presented with a different set of numbers that would have meant the earlier signal was not a signal to trade, or that there was a signal to trade shown, that didn't feature in the earlier calculations done in a live market setting. Faced with that conundrum I suspended trading Strategy II.

 

Whilst the underlying principle I was seeking to exploit with these strategies was essentially the same, each strategy was implemented in a sufficiently different way such that they were effectively two separate strategies.

 

I went back Niederhoffer's book, 'Practical Speculation' and recalled the importance of being able to quantify, whether it be a correlation, an edge, or apparent observation.

 

Using past signals for strategy II, I assessed these against short term market moves in Dow Jones futures. The way I opted to do this was to contrast the signal strength versus the immediate market activity following the signal, to try to establish whether or not an edge exists.

 

Normative_zps1aca8ebb.png

 

N on the x axis represents the signal, the stronger the signal is, the further to to the right it is shown on the scatter plot. The y axis shows the % move in Dow Jones futures, following a signal. So the bigger the percentage move, the further to the top of the scatter plot it is shown. I've then inserted a trend line that best describes the relationship between the two. The trend line is rising to the right, what this means is that we can say that based on the data set used (covering a few weeks of data), the stronger the signal, the bigger the move in Dow Jones futures. So there is a positive relationship between a signal, and a move in the underlying market. The fact that the swarm of dots do not fit very closely to the trend line is illustrative that the relationship is not a partcularly strong one, but it is there. The correlation co-efficient is 0.23.

 

 

Further to this I have been getting deeper into looking at different aspects of the underlying data and appear to have found some interesting results that if confirmed, provide a clear explanation as so why Strategy I broke down.

 

Rather than looking at the relationship between what I would class as being a signal, I opted to check if there is a relationship between the underlying data sets that I combine (lets call that N) in order to find signals, and the actual market prices, and what I have found may be valuable to explore further.

 

I looked at the relationship on individual days and what I found was that it changed from day to day. My previous understanding was that if N increased, the probability was that Dow Jones futures would increase. This meant that some days I was trading in a manner that was essentially the opposite to what I should have been doing.

 

Day1_zps6bdf0623.png

 

This scatter is particularly revealing to me since the overall relationship between N and the price of Dow Jones futures was one where if N increased, the price went down, although importantly, it's clear that there were periods within this day in questions where when N increased the price also increased, but these were smaller sub-relationships within the wider picture (show by areas where the dots form lines across the main trend, perpendicular to it).

 

I picked two other random days;

 

Day2_zps6ad7e9e0.png

 

Again, it's clear that there is a similar relationship here.

 

And another day I picked at random to assess;

 

Day3_zps359d7ade.png

 

On this day the relationship was completely different, whereby as N increased, market prices also increased, the opposite of the two other days shown.

 

What is particularly interesting in these example is that even without the trendlines showing the given relationship on each of these days, it would be clear what the relationship was without it, and moreover that the relationship didn't change during the day, at least in these example that I looked at. This is important because these are not time bound charts, therefore what it may be possible to do is asess the first period of any given trading day, perhaps 2 or 3 hours, to quantify the relationship on a given day, and then trade accordingly.

 

This is a potentially interesting development, but I need to do a lot more research on it.

Share this post


Link to post
Share on other sites

 

Thanks for posting, having had a brief scan through that looks very interesting, do you know anything about the background of the author, or have you read any of his other books, and how is your own research and trading system development going?

 

I own several of his books. In my view it is probably the best material I've come across regarding system development (testing, taking the system live, and ongoing system maintenance etc. - at least for EOD systems). Also, his website is full of useful links: http://www.blueowlpress.com/WordPress/links/

 

As for his background: he used to be a professor of mathematics & computing and did a stint at a hedge fund. The examples he develops are all end of day systems, but obviously one can translate the basic concepts to an intra-day time frame if so desired.

 

His platform of choice is AmiBroker, although he also uses Excel (for system monitoring and advanced analysis) and Python (to incorporate machine learning techniques into the system). His favoured strategy is counter trend with a short holding period (one to several days for EOD systems) trading liquid ETFs or large cap stocks; thereby minimising exposure, maximising the power of compounding, as well as being easier to deal with psychologically (i.e., aiming for systems with a 65%+ win/loss ratio and 1%+ expectancy - not unlike some of Larry Connors' strategies). And last but not least, it's easier to monitor systems that trade frequently.

 

More:

 

Chapter 2 - Mean Reversion Trading Systems, by Howard B. Bandy

http://www.meanreversiontradingsystems.com/MRTS%20AnalysisWM.pdf

 

MTA Webinar

http://www.blueowlpress.com/WordPress/2014/08/a-recording-of-the-quantitative-technical-analysis-webinar-is-available/

 

- - - -

 

On a related note, I recently also came across another useful website 'QuantStart' that is geared more towards the HFT end of the spectrum. Also, loads of useful links and articles.

 

http://www.quantstart.com/articles/Installing-a-Desktop-Algorithmic-Trading-Research-Environment-using-Ubuntu-Linux-and-Python

Share this post


Link to post
Share on other sites

I've been doing some really interesting work recently in an effort to define an edge within the market. Through the middle of last year to the start of this year I had been successfully trading a strategy (Strategy I) using an edge that I thought I had developed based on an underlying principle I had formed on some factors that influence short term price movement in stock index futures markets. That strategy broke down and instead of having a majority positive profitable days, it changed to a majority of negative losing days, therefore I had no option but to stop trading it. I never managed to understand why that happened, however based on recent work I think I may have now found the answer.

 

I also had traded a further strategy in recent weeks (Strategy II), which looked highly promising however calculation issues meant that I could take a signal to trade, find it to be a losing trade, then recalculate the data at day end to find myself presented with a different set of numbers that would have meant the earlier signal was not a signal to trade, or that there was a signal to trade shown, that didn't feature in the earlier calculations done in a live market setting. Faced with that conundrum I suspended trading Strategy II.

 

Whilst the underlying principle I was seeking to exploit with these strategies was essentially the same, each strategy was implemented in a sufficiently different way such that they were effectively two separate strategies.

 

I went back Niederhoffer's book, 'Practical Speculation' and recalled the importance of being able to quantify, whether it be a correlation, an edge, or apparent observation.

 

Using past signals for strategy II, I assessed these against short term market moves in Dow Jones futures. The way I opted to do this was to contrast the signal strength versus the immediate market activity following the signal, to try to establish whether or not an edge exists.

 

Normative_zps1aca8ebb.png

 

N on the x axis represents the signal, the stronger the signal is, the further to to the right it is shown on the scatter plot. The y axis shows the % move in Dow Jones futures, following a signal. So the bigger the percentage move, the further to the top of the scatter plot it is shown. I've then inserted a trend line that best describes the relationship between the two. The trend line is rising to the right, what this means is that we can say that based on the data set used (covering a few weeks of data), the stronger the signal, the bigger the move in Dow Jones futures. So there is a positive relationship between a signal, and a move in the underlying market. The fact that the swarm of dots do not fit very closely to the trend line is illustrative that the relationship is not a partcularly strong one, but it is there. The correlation co-efficient is 0.23.

 

 

Further to this I have been getting deeper into looking at different aspects of the underlying data and appear to have found some interesting results that if confirmed, provide a clear explanation as so why Strategy I broke down.

 

Rather than looking at the relationship between what I would class as being a signal, I opted to check if there is a relationship between the underlying data sets that I combine (lets call that N) in order to find signals, and the actual market prices, and what I have found may be valuable to explore further.

 

I looked at the relationship on individual days and what I found was that it changed from day to day. My previous understanding was that if N increased, the probability was that Dow Jones futures would increase. This meant that some days I was trading in a manner that was essentially the opposite to what I should have been doing.

 

Day1_zps6bdf0623.png

 

This scatter is particularly revealing to me since the overall relationship between N and the price of Dow Jones futures was one where if N increased, the price went down, although importantly, it's clear that there were periods within this day in questions where when N increased the price also increased, but these were smaller sub-relationships within the wider picture (show by areas where the dots form lines across the main trend, perpendicular to it).

 

I picked two other random days;

 

Day2_zps6ad7e9e0.png

 

Again, it's clear that there is a similar relationship here.

 

And another day I picked at random to assess;

 

Day3_zps359d7ade.png

 

On this day the relationship was completely different, whereby as N increased, market prices also increased, the opposite of the two other days shown.

 

What is particularly interesting in these example is that even without the trendlines showing the given relationship on each of these days, it would be clear what the relationship was without it, and moreover that the relationship didn't change during the day, at least in these example that I looked at. This is important because these are not time bound charts, therefore what it may be possible to do is asess the first period of any given trading day, perhaps 2 or 3 hours, to quantify the relationship on a given day, and then trade accordingly.

 

This is a potentially interesting development, but I need to do a lot more research on it.

 

That looks promising!

 

I've mainly been developing EOD systems, but have already encountered some glitches (not unlike your strategy II problem) in the execution of the system. Nothing fatal, but the take home message is: it's one thing to backtest the data offline (in/out of sample), and quite another taking it online. It's funny because it's almost cliche, and it's not as if I was unaware of the potential for such issues, but it's always different when one encounters it live - in the flesh, so to speak.

Share this post


Link to post
Share on other sites

I've been really busy on a professional level recently (not a lot of time for posting) however have been able to work on this some more and think that I can use this type of research to inform trading decisions. There's more work to be done but what is quite interesting is that relationships between market data are assessed, this then defines what is likely to constitute a signal on a daily basis, so what may have been a signal one day may not be the next, in that sense it is dynamic and changes dependent on market conditions.

 

All my previous strategies have been linear, not responding to the market but rather trying to force my ideas on to the market, irrespective of market conditions, which doesn't work, a two way dialogue appears to be the way to go.

 

Anyway I have more time in the next two weeks to work on it.

Share this post


Link to post
Share on other sites

Pearson_zps06cf5984.png

 

I've been doing a lot of research on factors that taken together correlate with the price of Dow Jones 30.futures. The chart above shows the 4 factors (denoted N on the chart) and their relationship to the change in price of Dow Jones futures over a set timeframe.So the relationship between the change in value of N to the change in value of Dow Jones 30. I'm using the Pearson product -moment correlation coefficient which is a measure of the linear correlation between two variables, and this research uses market data from 14th August 2014 until 10th October 2014. The Pearson correlation is 0.82 which is a strong correlation (Maximum is 1), what's interesting is the strength of the correlation over a large period of time using a lot of data points (over 60,000 x and y pairs), with quite a tight variance in terms of the % change in price.

 

Whether or not this work can be progressed to a tradeable insight is another matter and needs more research.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×