Laboratory of Structural Methods of Data Analysis in Predictive
Modeling Moscow Institute of Physics and Technology

Financial engineering and energy management

Описание проекта:

As opposed to purely financial obligations like stocks and bonds, energy (electricity, natural gas, oil etc.) can be physically stored. Storage allows for temporal transfer of the energy and permits exploitation of the fluctuating market energy prices. The basic principle here is to ‘buy low’ and ‘sell high’, such that the realized profit covers storage and operating costs. The opportunities for the profit depend on the dynamics of the price process and on the ability of the energy holder to forecast it. This results in high demand on statistical prediction algorithms in energy sector.

Usually major players hold storage facilities in the respective industries, since they had the sufficient capital needed to build and maintain them. However, with the liberalized markets, all participants have nowadays the opportunity to rent a storage facility to speculate on prices. Therefore it becomes necessary to compute the financial value of the storage facility. Namely, one can ask how much one can profit from getting control of a storage facility for a period of given number of years? Another related question is how to describe the optimal strategies for buying, storing and selling energy in time.

There is no simple answer to these questions, as the storage facility holder faces many operational and engineering constraints like inventory capacity limits or delivery charges which themselves depend on the chosen storing strategy. Another important problem is the accurate short time forecasting of energy prices, which is crucial for the construction of reliable selling, buying and storing strategies.

To properly account for the dependence between the timing optionality in choosing the purchase and sale times and the inventory constraints, one has to consider the full stochastic control framework. This leads to a Bellman dynamic programming equation for the value function. From here one may use the Hamilton-Jacobi-Bellman theory, recasting the problem into a quasi-variational partial differential equation (pde) formulation.

Another approach is simulation-based. It approximates the solution of the optimal stochastic control problem by using Monte Carlo methods combined with modern regression methods and methods of convex optimization. Monte Carlo methods are especially attractive here as the energy price process usually depends on many endogenous factors (season, weather and etc.) leading to high-dimensional control problems.

Члены лаборатории:
Dorn Yuriy
Должность: Младший научный сотрудник
Spokoiny Vladimir
Должность: Руководитель
Golubev Yuri
Должность: Ведущий научный сотрудник
Nesterov Yurii
Должность: Ведущий научный сотрудник

Должность: Ведущий научный сотрудник