Analysing Customer Behaviour in the FX Market Using Order Flow Data and Machine Learning Techniques
(2015) FMS820 20151Mathematical Statistics
- Abstract (Swedish)
- This thesis has two main objectives related to trading foreign currencies. First, it is
investigated how the customer order
ow of Nordea is related to currency price changes.
Second, the goal is to nd a new way of grouping customers that can give additional
insights in the trading behaviour of dierent customers.
The study uses order
ow data which consists of spot and forward transactions made
in Norwegian Kronor and Swedish Kronor during a period of nearly three years. The
counterparties with whom Nordea is trading foreign exchange are divided into the customer
groups asset managers, banks, corporates, hedge funds and private clients. As a
measure of the dependence between order
ow and exchange rate movements an index is
... (More) - This thesis has two main objectives related to trading foreign currencies. First, it is
investigated how the customer order
ow of Nordea is related to currency price changes.
Second, the goal is to nd a new way of grouping customers that can give additional
insights in the trading behaviour of dierent customers.
The study uses order
ow data which consists of spot and forward transactions made
in Norwegian Kronor and Swedish Kronor during a period of nearly three years. The
counterparties with whom Nordea is trading foreign exchange are divided into the customer
groups asset managers, banks, corporates, hedge funds and private clients. As a
measure of the dependence between order
ow and exchange rate movements an index is
introduced. The index tells how the spot exchange rate moves before and after a trade
is executed. To examine the trading behaviour for a customer group the indices for all
trades done by that group are weighted with the traded volume.
Grouping customers in a new way is addressed by using Machine Learning techniques
in the eld unsupervised learning, called clustering. The applied clustering algorithms
are the K-means, the Fuzzy C-means and the Self-Organizing Map. The customers are
clustered according to four dierent features calculated from the order
ow data.
The main ndings include that there are dierences in the dependence between exchange
rate changes and the order
ow from dierent customer groups. A contrarian behaviour
is found for the corporates and private clients, while the asset managers and hedge
funds tend to hold a trend-following trading style. The results from the clustering do
not provide a better way of grouping clients than the previous one, but contributes with
a deeper understanding of the trading behaviour of dierent customers. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/5043462
- author
- Thordin, Lovisa
- supervisor
- organization
- course
- FMS820 20151
- year
- 2015
- type
- H2 - Master's Degree (Two Years)
- subject
- language
- English
- id
- 5043462
- date added to LUP
- 2015-02-04 15:22:54
- date last changed
- 2015-02-04 15:22:54
@misc{5043462, abstract = {{This thesis has two main objectives related to trading foreign currencies. First, it is investigated how the customer order ow of Nordea is related to currency price changes. Second, the goal is to nd a new way of grouping customers that can give additional insights in the trading behaviour of dierent customers. The study uses order ow data which consists of spot and forward transactions made in Norwegian Kronor and Swedish Kronor during a period of nearly three years. The counterparties with whom Nordea is trading foreign exchange are divided into the customer groups asset managers, banks, corporates, hedge funds and private clients. As a measure of the dependence between order ow and exchange rate movements an index is introduced. The index tells how the spot exchange rate moves before and after a trade is executed. To examine the trading behaviour for a customer group the indices for all trades done by that group are weighted with the traded volume. Grouping customers in a new way is addressed by using Machine Learning techniques in the eld unsupervised learning, called clustering. The applied clustering algorithms are the K-means, the Fuzzy C-means and the Self-Organizing Map. The customers are clustered according to four dierent features calculated from the order ow data. The main ndings include that there are dierences in the dependence between exchange rate changes and the order ow from dierent customer groups. A contrarian behaviour is found for the corporates and private clients, while the asset managers and hedge funds tend to hold a trend-following trading style. The results from the clustering do not provide a better way of grouping clients than the previous one, but contributes with a deeper understanding of the trading behaviour of dierent customers.}}, author = {{Thordin, Lovisa}}, language = {{eng}}, note = {{Student Paper}}, title = {{Analysing Customer Behaviour in the FX Market Using Order Flow Data and Machine Learning Techniques}}, year = {{2015}}, }