Data Science Projects

Recommendations System for a Webshop
A client is a clothing online store that requested to create an algorithm that would recommend potentially interesting and relevant items for clients to buy before checkout.
Human Trajectories
A client is a WiFi provider company that requested to create an algorithm that would identify meetings between users in a shopping mall by tracking the WiFi signals on their devices.
Predictive model for oil and gas exploration data
A client is a startup that requested to build a model identifying a connection between various types of fracking techniques and fracking efficiency.
Transportation costs optimisation
A client is a online retailer that requested to optimise their packing procedures for minimising costs purposes.
Menu recommendation system
A client is a restaurant that requested to do a customer preferences analysis.

Research Projects

Empirical Analysis of the Power Price
We conduct an empirical analysis widely used models for electricity spot price process. Further we calibrate all three models to German spot price data. Besides employing techniques similar to those used in the original models we adopt some new statistical techniques. We critically compare the properties and the estimation of the three models and discuss several shortcomings and possible improvements. Besides analysing the spot price behaviour, we compute forward prices and risk premia for all three models for various German forward data and identify the key forward price drivers.
Model Risk in Energy Markets
This is a joint project with the research group from TU Munchen, particularly with Karl Bannoer and Mathias Scherer. In the financial world, there has been developed a rich zoo of different models which can be used for financial purposes. But one cannot be sure that the model is always suitable and, more microstructural, whether the model's parameters are correctly obtained. In some cases, during the financial crisis, the choice of not-suitable models has been isolated as one of the major sources of the distress for banks, e.g. the use of the Gaussian copula model. Hence, one is typically exposed to model uncertainty. In some cases, one might be able to assign probabilities to the different models, where one ends up with model risk. Recently, a number of authors addressed this issue. In contrast, model risk has not been discussed in the context of energy markets. However, taking into account the recent changes in the European Energy Market, in particular the German "Energiewende" it is clear that model risk is a pressing issue. On particular feature is the need for reinvestment (replacement investments and building more capacity) in the power plant park on Company and European level. The financial streams of such an investment can be generated on the market for energy derivatives in terms of spread options. For instance, a gas-fired power plants can be represented as a clean crack spread option, where the owner of such an option is long electricity and short gas and emission certificates. A positive investment decision is made in case such a contract is in the money, meaning that we observe a positive spread on the time interval under consideration. In terms of this project We aim to analyse the model risk inherent in such an investment decision.
State-dependent Jumps in the Forward Dynamics
We are focused in the model with jumps which are frequently observed for electricity spot prices, but are state-dependent. The question we are interested in is to understand the forward price dynamics.
Storage Modelling
We suggest a new approach to model storage level process. Instead of widely used stochastic optimal control approach, we assume that we already have an optimal storage policy and investigate various payoffs that help to hedge the market position for a power producer.
Diffusion with Delayed Reflection
This project investigates the diffusion which has a delayed reflection and lives between two boundaries. First, we specify the diffusion coefficients. Then by using Green's functions method we derive the transition density for such process. This case can be viewed as the development of the case given in the book "Stochastic Differential Equations" by I. I. Gihmand and A. V. Skorokohod, 1972. The application of such process might be possible in the area of carbon markets, where the CO2 price has floor and ceiling. Also modeling the fuel storage capacity can be considered as well.