|
|
Dynamic decision making based on iterations-spread-in-time strategy
Šindelář, Jan ; Křivánek, O.
This article describes a formal approach to decision making optimization in commodity futures markets. Our aim was to design optimal decision strategy generating decision at a given time. It contains theoretical description of estimation using Bayesian learning and approximate methods of dynamic programming. Finally, the original decision strategy using approximate methods of dynamic programming was designed. This strategy was tested by a series of experiments indicating our ability to construct pro table trading machine.
|
|
|
Transformace finančních dat určených pro dynamické rozhodování
Chudoba, M. ; Jirsa, Ladislav
In the presented work we are introduced to the problem of optimal decision making while dealing on the exchange with so-called "financial futures", i.e. time financial transaction. This task is transferred into the simplified mathematical model, which is solvable using Bayesian estimation methods. Financial data are modelled by auto-regressive model with normal noise, because the tools, which are exploited for prediction of the price on the market and which assume normal noise, have been already developed. The main goal of this work is the comparison of the efficiency of various transformations on input data, so that their noise had normal distribution, therefore the price prediction was as accurate as possible. The applicable algorithm is programmed in Matlab; the presentation of achieved results forms the final part of this thesis.
|
|
|
Řízení výzkumné skupiny pomocí odměn
Kárný, Miroslav
This text was stimulated by the need to distribute motivating personal money between a group of researchers. These money should primarily stimulate an increase the group productivity. Attempts to distribute the money in a fair way that respects their simulating aspect have led to the conclusion that the task is a difficult dynamic decision problem under uncertainty. The basic solution idea is straightforward. Significant groups of measurable outputs of research work are scaled to the common financial basis. Personal money are then distributed proportionally to he latest individual results.
|
|
|
Adaptivní dopředné řízení
Kárný, Miroslav
Feed-forward controllers are important in a range of control problems. Their importance is obvious in tasks in which potential of feedback control is limited, for instance, in systems with long transportation delays. Formally, their design can be approached by a standard methodology of optimal stochastic control. It can be consistently performed by combining Bayesian learning and dynamic programming. This formal solution can, however, rarely be converted into a computationally feasible algorithms. Thus, various approximations are searched for. The current report deals with a specific type of approximation based on a projection of optimal /emph{anticipating} control strategy to a non-anticipating one. This approximation way suits to the feed-forward control in which the selected system inputs influence the state of the controller but not the system-related data used in the feed-forward loop.
|
|
| |
|
|
Adaptivní řízení aplikované na data z finančních trhů
Šindelář, Jan ; Kárný, Miroslav
The article describes a formal approach to decision making optimization in commodity futures markets. We try to plan optimal decision at a given time to trade in the market. We use dynamic programming with loss function equal to the negative profit, where we estimate the PDFs of parameters using Bayesian learning. Parametrized models are chosen from exponential family and trading costs (slippage and commission) are taken into account. We support the theoretical results by a series of experiments.
|
|
|
Spojování znalostí jako problém odhadování
Kárný, Miroslav ; Bodini, A. ; Ruggeri, F.
Bayesian cooperative decision making by multiple participants requires a combination of environment models of cooperating neighbors. In a sense, a speci¯c participant tries to exploit imprecise probabilistic knowledge con- cerning of low-dimensional marginal and conditional distributions o®ered by his neighbors. In spite of the substantial overlap with domains like knowl- edge elicitation, estimation of graphical models or building of probabilistic expert systems, the merging problem was not addressed yet in its entirety.
|
|
|
Příklady odhadu stavu a parametrů pro lineární model s rovnoměrně rozloženými inovacemi
Pavelková, Lenka
In this contribution, state-space model with uniformly distributed innovations is introduced and the Bayesian state estimation proposed. The off-line evaluation of the maximum a posteriori probability (MAP) estimate of unknowns in the linear state-space model with uniform innovations reduces to linear programming (LP). The solution provides either estimates of the noise boundary and parameters or of the noise boundary and states. The on-line estimation is obtained by applying LP on the sliding window, i.e., by considering only the fixed amount, say partial, of the newest last data and states items. By swapping between state and parameter estimations, joint parameter and state estimation is obtained. The use of Taylor expansion for approximation of products of unknowns solves also the joint parameter and state estimation. Simulation studies help to get an insight on the potential and restrictions of these heuristic method. This contribution shares the experimentally gained experience with both these solutions of the joint state and parameter estimation.
|
|
| |
guest ::
