EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
555. A reinforcement learning mechanism for trading wind power futures
Invited abstract in session TD-18: AI and ESG for the small economy SDG agenda (EWG-ORD Workshop 2), stream OR for Development and Developing Countries.
Tuesday, 14:30-16:00Room: 42 (building: 116)
Authors (first author is the speaker)
| 1. | Bruno Kamdem
|
| Department of Business Management, SUNY Farmingdale State College, School of Business | |
| 2. | Moustapha Pemy
|
| Mathematics, Towson University |
Abstract
Recent developments in Europe have disrupted gas supplies and reignited the urgency of moving towards electricity production from winds. But, the narrow predictability of wind and the intermittency of electricity generation can complicate
decisions within trading desks. To alleviate the issue at hand, an attempt in this paper is to provide an optimal training mechanism by accounting seasonalities in wind productions as electricity sells in the exchange. We formulate the problem from a standpoint of an institutional investor concerned with ESG factors. For the trader, we define his reward at each
future price after a number of trades. We introduce a parametrized approximate value function using deep convolutional neural network in which we specify the weights of the Q-network at each iteration. We derive some key implications for the trader as to the best time to buy wind power futures and at what price to do it.
Keywords
- Machine Learning
- Environmental Management
- Financial Modelling
Status: accepted
Back to the list of papers