Date post: | 20-Dec-2015 |
Category: |
Documents |
View: | 215 times |
Download: | 1 times |
11
Indirect Estimation of the Indirect Estimation of the Parameters of Agent Based Parameters of Agent Based Models of Financial MarketsModels of Financial Markets
Peter WinkerPeter Winker
Manfred GilliManfred Gilli
22
OutlineOutline
Background Information.Background Information. Introduction.Introduction. Method.Method. Result.Result. Conclusion of the Paper.Conclusion of the Paper. Further Improvement.Further Improvement.
33
Standard ModelsStandard Models
AssumptionsAssumptions Agents are fully rational.Agents are fully rational. Markets are efficient.Markets are efficient.
44
Rational BehaviourRational Behaviour
Agent is rational ifAgent is rational if He is aware of his alternatives.He is aware of his alternatives. Form expectations about any unknowns.Form expectations about any unknowns. Has clear preferences.Has clear preferences. Chooses his action deliberately after Chooses his action deliberately after
some process of optimization.some process of optimization. Taking into account their knowledge or Taking into account their knowledge or
expectations of other decision makers’ expectations of other decision makers’ behaviour.behaviour.
55
Efficient Market HypothesisEfficient Market Hypothesis
All market participants receive and All market participants receive and act on all relevant information as act on all relevant information as soon as it is available.soon as it is available.
Perfect information within the Perfect information within the market.market.
Cannot “beat the market”.Cannot “beat the market”.
66
Agent Based ModelsAgent Based Models
Agents to be heterogenous.Agents to be heterogenous. Agents with limited rational Agents with limited rational
behaviour.behaviour. Market does not need to be efficient.Market does not need to be efficient. Interaction between agents.Interaction between agents.
77
ParametersParameters
Not directly observable.Not directly observable. Compare with empirical data.Compare with empirical data. DM/US-$ exchange rate.DM/US-$ exchange rate.
88
Characteristic of DM/US-$Characteristic of DM/US-$
Daily Returns DM/US-$Daily Returns DM/US-$
99
Characteristic of DM/US-$Characteristic of DM/US-$
Excess kurtosis.Excess kurtosis.
Volatility varies over time.Volatility varies over time. AR(1) process with ARCH(1) effect.AR(1) process with ARCH(1) effect. where where
4XXE
ttt rr 110 2
110 ttV
1010
Model (Kirman 1990)Model (Kirman 1990)
Two prevalent views of the world.Two prevalent views of the world. Each agent holds one view.Each agent holds one view. NN agents. agents. State: number of agents, State: number of agents, kk, for first view., for first view. Two agents, A and B, meet at random.Two agents, A and B, meet at random. P(A’s view P(A’s view →→ B’s view) = B’s view) = .. P(A changed his view independently) = . P(A changed his view independently) = . 1
1111
Model (Kirman 1990)Model (Kirman 1990)
If , large shares of first type of If , large shares of first type of agents and second type of agents, agents and second type of agents, respectively, with high probability. respectively, with high probability.
111,
N
k
N
kNkkP
1
11,N
kN
N
kkkP
1
1
N
1212
Fundamentalist / ChartistsFundamentalist / Chartists
There are two types of agents:There are two types of agents: Fundamentalist:Fundamentalist:
Chartist:Chartist:
ttf SSvSE 1
11 tttc SSSE
1313
AdvantagesAdvantages
Complicated non-stationary Complicated non-stationary dynamics.dynamics.
Non-fundamentalist behaviour.Non-fundamentalist behaviour.
1414
SimulationSimulation
Objective function.Objective function. estimated ARCH(1)-effect.estimated ARCH(1)-effect. empirical kurtosis.empirical kurtosis. and mean values from 1000 and mean values from 1000
simulations.simulations. First and last 10% results deleted.First and last 10% results deleted.
empempdd kkf 11
emp1
empdk
dk 1
1515
Monte Carlo SimulationMonte Carlo Simulation
200 Monte Carlo simulation200 Monte Carlo simulation
1616
Monte Carlo SimulationMonte Carlo Simulation
10000 Monte Carlo simulation10000 Monte Carlo simulation
1717
Monte Carlo SimulationMonte Carlo Simulation
10000 Monte Carlo simulation10000 Monte Carlo simulation
1818
Threshold AcceptingThreshold Accepting
Initial:Initial: Choose threshold sequence Choose threshold sequence
, set, set
and generate an initial .and generate an initial . Step 1:Step 1: Choose some .Choose some . Step 2:Step 2: Calculate .Calculate . Step 3:Step 3: If , set .If , set . Step 4:Step 4: If , set and go to 1.If , set and go to 1.
Otherwise, output .Otherwise, output .
max,,0, IiTi cx0i
cn xNx cn xfxff
iTf nc xx
maxIi 1iicx
1919
SimulationSimulation
2020
SimulationSimulation
2121
ResultResult
Optimal values are andOptimal values are and
, market is better , market is better characterized by switching moods of characterized by switching moods of the investors than by assuming that the investors than by assuming that the mix of fundamentalists and the mix of fundamentalists and chartists remains rather stable over chartists remains rather stable over time.time.
0008571.03250.0
1
1
N
2222
ConclusionConclusion
Agent based models can replicate Agent based models can replicate empirical data of the financial empirical data of the financial markets.markets.
Parameters may be difficult to Parameters may be difficult to estimate.estimate.
Indirect method can be used.Indirect method can be used. Optimization heuristic may need to be Optimization heuristic may need to be
used.used.
2323
Further ImprovementFurther Improvement
First and last 10% simulation results First and last 10% simulation results removed. Too much?removed. Too much?
Number of parameters to be Number of parameters to be estimated.estimated.
Only two types of agents?Only two types of agents?
2424
ReferenceReference
Fama, E.F. 1970, “Efficient capital markets: a review of theory and Fama, E.F. 1970, “Efficient capital markets: a review of theory and empirical work”, empirical work”, Journal of FinanceJournal of Finance, V25, Issue 2, p383-417., V25, Issue 2, p383-417.
Gilli, M., Winker, P. 2003, “A global optimization heuristic for Gilli, M., Winker, P. 2003, “A global optimization heuristic for estimating agent based models”, estimating agent based models”, Computational Statistics & Data Computational Statistics & Data AnalysisAnalysis, 42, p299-312., 42, p299-312.
Kirman, A. 1990, “Epidemics of opinion and speculative bubbles in Kirman, A. 1990, “Epidemics of opinion and speculative bubbles in financial markets”, in Taylor M.P.(eds), financial markets”, in Taylor M.P.(eds), Money and financial Money and financial marketsmarkets, Basil Blackwell Ltd, Oxford, p354-368., Basil Blackwell Ltd, Oxford, p354-368.
Tsay, R.S. 2002, Tsay, R.S. 2002, Analysis of financial time seriesAnalysis of financial time series, John Wiley & , John Wiley & Sons, Inc.Sons, Inc.
Winker, P. 2001, “Application of the optimization heuristic Winker, P. 2001, “Application of the optimization heuristic threshold accepting in statistics”.threshold accepting in statistics”.