Advanced

Tick data clustering analysis establishing support and resistance levels of the EUR-USD exchange market

Tengelin, Karl LU (2020) In Master’s Theses in Mathematical Sciences FMSM01 20201
Mathematical Statistics
Abstract
Our aim is to use clustering algorithms in order to compute support and resistance levels within an intra-day trading setting. To achieve this we use a tick data set from the EUR-USD exchange market during 2019 as a measure of market activity. Both the Gaussian Mixed Model (GMM) and an altered form of Kmeans clustering will be used as clustering methods where each method will be evaluated using a selection of common performance metrics. The computed support and resistance levels will then be put to the test by initiating mock trades during certain time windows from early 2019, which are specified by Century Analytics.

Both models that were used in this thesis managed to partition the data in a way that made it possible to create support... (More)
Our aim is to use clustering algorithms in order to compute support and resistance levels within an intra-day trading setting. To achieve this we use a tick data set from the EUR-USD exchange market during 2019 as a measure of market activity. Both the Gaussian Mixed Model (GMM) and an altered form of Kmeans clustering will be used as clustering methods where each method will be evaluated using a selection of common performance metrics. The computed support and resistance levels will then be put to the test by initiating mock trades during certain time windows from early 2019, which are specified by Century Analytics.

Both models that were used in this thesis managed to partition the data in a way that made it possible to create support and resistance levels that are comparable to traditional methods which do not rely on market activity. Although more research needs to be made the results look promising and we can, with some confidence, say that market activity in the form of ticks can be used as an indicator for support and resistance levels within the EUR-USD exchange market.

The support and resistance levels computed using GMM and Kmeans were quite similar but the GMM method performed better when examining the methods using mock trades. The GMM could predict support and resistance ”bounces” with greater statistical significance compared to the Kmeans method. (Less)
Popular Abstract
Trading has become a far more intricate and data-driven business today
compared to a few years ago and companies in the financial sector constantly
strive to get an edge over the competition. Hence it is only natural that these
companies have become more interested in Machine Learning and AI, which are
groups of methods who are able to process large data sets efficiently. Therefore
we wanted to examine if we can use Machine Learning methods coupled with a
relatively known indicator of price - market activity - in order to increase the
probability of trading with profit. In other words, how do we make better data
driven decisions using Machine Learning and market activity?

We investigated this by looking at tick data gathered... (More)
Trading has become a far more intricate and data-driven business today
compared to a few years ago and companies in the financial sector constantly
strive to get an edge over the competition. Hence it is only natural that these
companies have become more interested in Machine Learning and AI, which are
groups of methods who are able to process large data sets efficiently. Therefore
we wanted to examine if we can use Machine Learning methods coupled with a
relatively known indicator of price - market activity - in order to increase the
probability of trading with profit. In other words, how do we make better data
driven decisions using Machine Learning and market activity?

We investigated this by looking at tick data gathered from the EUR-USD
exchange market, which was provided by Century Analytics, with the aim to
use this data set to compute support and resistance levels. These are levels
where the price trend, due to market psychology, tends to change direction. If
we could predict these changes in direction we could in theory invest over short
time spans with very low risk.

Another common concept in trading aside from support and resistance levels
is market activity. Market activity is commonly used by financial institutions
and companies when trying to analyse the market. In this case market activity
refers to how many orders to either buy or sell are placed on the EUR-USD
exchange market each fraction of a second (known as ticks). The underlying
assumption here is that market activity and the variance of the price correlates,
which we could confirm.

If we combine these two elements, support/resistance levels and market activity, we have the two cornerstones on which this project was built. Next step
is to introduce machine learning into the equation.

Using a Gaussian Mixed Model (GMM) with Expectation Maximisation and
an altered form of Kmeans combined with a selection of cluster performance
evaluations we were able to partition the tick data into clusters that could be
used to establish support and resistance levels for time series data of EURUSD exchange rates. This means that the algorithms were able to predict price
changes at least to some degree without receiving any information about time
series data of the exchange rates, but rather using market activity as the only
indicator of price changes. The computed support and resistance levels were
then tested using mock trades to investigate the consistency of the methods
and to see if we could claim any statistical significance to our findings.

Our best results came from using a GMM algorithm which found support
and resistance levels around 4 percentage points more often than not, meaning
that the algorithm was right 4 percentage points more often than it was wrong.
This means that using this method to trade during the time span we investigated
would actually result in a profit.

When testing for statistical significance we did however find that we cannot
really reject the null hypothesis that the findings are random, hence further
investigation is probably needed before one could motivate risking money on this
method. We did however see that the GMM method produced some statistically
significant results over some intervals of the provided tick data, something that
motivates further studies on the subject. (Less)
Please use this url to cite or link to this publication:
author
Tengelin, Karl LU
supervisor
organization
alternative title
Klustringsanalys av tickdata i syfte att fastställa support-och resistansnivåer för valutahandeln av EUR-USD
course
FMSM01 20201
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Tick data, Support-and resistance levels, Clustering methods, Gaussian mixture model, Kmeans, EUR-USD exchange rates, Clustering performance metrics, Market activity
publication/series
Master’s Theses in Mathematical Sciences
report number
2020:E24
ISSN
1404-6342
language
English
id
9008772
date added to LUP
2020-05-13 07:54:33
date last changed
2020-05-13 07:54:33
@misc{9008772,
  abstract     = {Our aim is to use clustering algorithms in order to compute support and resistance levels within an intra-day trading setting. To achieve this we use a tick data set from the EUR-USD exchange market during 2019 as a measure of market activity. Both the Gaussian Mixed Model (GMM) and an altered form of Kmeans clustering will be used as clustering methods where each method will be evaluated using a selection of common performance metrics. The computed support and resistance levels will then be put to the test by initiating mock trades during certain time windows from early 2019, which are specified by Century Analytics. 

Both models that were used in this thesis managed to partition the data in a way that made it possible to create support and resistance levels that are comparable to traditional methods which do not rely on market activity. Although more research needs to be made the results look promising and we can, with some confidence, say that market activity in the form of ticks can be used as an indicator for support and resistance levels within the EUR-USD exchange market.

The support and resistance levels computed using GMM and Kmeans were quite similar but the GMM method performed better when examining the methods using mock trades. The GMM could predict support and resistance ”bounces” with greater statistical significance compared to the Kmeans method.},
  author       = {Tengelin, Karl},
  issn         = {1404-6342},
  keyword      = {Tick data,Support-and resistance levels,Clustering methods,Gaussian mixture model,Kmeans,EUR-USD exchange rates,Clustering performance metrics,Market activity},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master’s Theses in Mathematical Sciences},
  title        = {Tick data clustering analysis establishing support and resistance levels of the EUR-USD exchange market},
  year         = {2020},
}