dr. T.A. (Tomasz) Makarewicz MSc

Faculty of Economics and Business
Section Quantitative Economics

Visiting address
REC E
Roetersstraat 11 Amsterdam
Room number: 3.53

Postal address:
Postbus 15867
1001 NJ Amsterdam

T.A.Makarewicz@uva.nl
T: 0205258766
Office hours
Tuesdays, 14:0017:00, room 2.08. Appointment via email possible.
Curriculum Vitae
Education
MA in Economics, CEU (2008)
MA in International Economics, Warsaw University (2008)
MA in Philosophy, Warsaw University (2009)
MPhil in Economics, Tinbergen Institute (2011)
Research interests
Evolutionary economics
Bounded rationality and learning
Agentbased modelling
Genetic algorithms
Libraries and programs
On this webpage you can find the libraries I have been using for my research. Each comes as a zip file with basic code, possibly a manual and ReadMe.txt file for installation notes.
You are free to download and use my libraries under the following terms of use:
The libraries can be freely used for academic, teaching or research purposes, provided that the authors quote attached manual and this webpage. The library cannot be used for any commercial purposes. Reader is free to change a library any way he or she desires, but if he or she wants to publish these changes, he or she should modify the versioncomment in the respective file.
Genetic Algorithms
Below is the link to the GA library I have been using during my phd. It requires Ox, with a free version available at Jan Doornik's homepage.The .zip file contains .h file with the class, three templates and a manual about using the class.
GA class: GAClass.zip.
PhD Lunch Seminar 2012/2013
Introduction
On this webpage you will find information about CeNDEF 2012/13 PhD Lunch Seminarseries in which our Phd students bring forward the topics of their current (or past) research.
The goal of the seminar is to provide communication between the young researchers and the academic stuff. In this way, students have an excellent opportunity to receive important feedback and discussion of their work from other students and professors, while professors can learn the interests and performance of CeNDEF students.
Lunch with tea or coffee is provided.
Time and place: Mon 13001400, 2.50/JK
The seminar takes place almost every Monday, at 13001400 in room 2.50 in the J/K builiding, CeNDEF, University of Amsterdam, Valckenierstraat 6567, 1018 XE Amsterdam, The Netherlands.
Contact information
For additional information, please email D.Kopanyi@uva.nl or T.A.Makarewicz@uva.nl orf.j.t.sniekers@uva.nl .
Timetable
Upcoming:
13th November
Speaker: David Kopanyi (CeNDEF UvA)
Title: PriceQuantity Competition with riskaverse firms
Abstract: I consider a market of a homogeneous good in which firms
simultaneously decide on both the price and produced quantity of their good. It
is known from the literature that there exists no purestrategy Nash equilibrium
in this setup and that there exists a unique mixedstrategy Nash equilibrium
when the demand function is linear and the marginal cost is constant. Each firm
chooses a price from a certain distribution and it produces the market demand at
that price in this mixedstrategy Nash equilibrium.
In laboratory experiments subjects typically produce less than the market demand
at their chosen price. My conjecture is that this can be explained with risk
aversion. Subjects can be betteroff with a lower production if they don't have
the lowest price. If they do have the lowest price, however, then they should
produce up to the market demand. However, they don't know in advance whether
they will have the lowest price or not. Therefore, there is a tradeoff in the
quantity choice. I want to analyze this question formally, assuming that firms
perform a mean  variance maximization for making a decision. Furthermore, I
want to propose a model in which firms use simple rulesofthumbs for making the
pricequantity choice that can explain the outcomes of laboratory experiments.
PhD Lunch Seminar 2011/12
Introduction
On this webpage you will find information about CeNDEF 2011/12 PhD Lunch Seminarseries in which our Phd students bring forward the topics of their current (or past) research.
The goal of the seminar is to provide communication between the young researchers and the academic stuff. In thisway, students have an excellent opportunity to receive important feedback and discussion of their work from other students and professors, while professors can learn the interests and performance of CeNDEF students.
Lunch with tea or coffee is provided.
Time and place: Tue 12001300, 2.08/JK
The seminar takes place almost every Tuesday, at 12001300 in room 2.50 in the J/K builiding, CeNDEF, Department of Quantitative Economics, University of Amsterdam, Valckenierstraat 6567, 1018 XE Amsterdam, The Netherlands.
Contact information
For additional information, please email M.Wolski@uva.nl or T.A.Makarewicz@uva.nl.
Timetable
14th February
Speaker: Domenico Massaro (CeNDEF UvA)
Abstract: In this paper we develop and estimate a behavioral model of inflation dynamics with monopolistic competition, staggered price setting and heterogeneous firms.
In our stylized framework there are two groups of price setters, fundamentalists and naive. Fundamentalists are forwardlooking and believe in a presentvalue relationship between inflation and real marginal costs, while naive are backwardlooking, using the simplest rule of thumb, naive expectations, to forecast future inflation. Agents are allowed to switch between these different forecasting strategies conditional on their recent relative forecasting performance.
The estimation results support behavioral heterogeneity and the evolutionary switching mechanism. We show that there is substantial time variation in the weights of forwardlooking and backwardlooking behavior. Although on average the majority of firms use the simple backwardlooking rule, the market has phases in which it is dominated by either the fundamentalists or the naive agents.
Seminar cancelled.
Speaker: Daan in 't Veld (CeNDEF UvA)
Title: Finding the Core: Network structure in interbank markets
Abstract: This paper investigates the network structure of interbank markets. We reproduce a recent analysis of Craig and Von Peter (2010), now using interbank exposure data in the Netherlands. We show that the Core Periphery model, previously found to be a very good description of the German interbank market, has somewhat less explanatory power for the equivalent Dutch market. We test for significance by comparing the outcomes for the Dutch data with random benchmark networks. The analysis opens up new opportunities for systemic risk assessments of the interbank market. (Joint with Iman van Lelyveld of DNB.)
Speaker: Aad Ruiter (ING)
Title: Price Discovery with Fallible Choice
Abstract: Experimental economics has shown that competitive equilibrium can be achieved under conditions which are mild compared to the `heroic assumptions' of general equilibrium theory. Simulation research may help to establish the wider conditions under which competitive equilibrium may obtain. To that end, define `fallible choice' as boundedly rational behavior, vulnerable to error and bias, which is sufficiently well adapted to its environment so as to survive competition with rational behavior. To determine (i) whether such viable fallible choice exists, and (ii) if it can achieve competitive equilibrium in an exchange economy, I define a number of cases which can and (later) will be simulated. The cases carry ideas of behavioral economics, of mental accounting in particular, over to the context of general equilibrium theory. The static equilibria, which they induce, already look interesting because some of them allow rationing even though prices are flexible.
Speaker: Xindan Li (Nanjing University)
Title: The impacts of algorithmic trading on execution cost, market quality, and trading system  real datadriven multiagent simulation of financial market
Abstract: This paper builds a Datadriven MultiAgent model to investigate the impact of algorithmic trading on execution costs, market quality, and trading system. The approach integrates order book information from real world with a simulation of an artificial financial market. We find that (1) The average execution costs of VWAP and IS algorithm are lower than the actual transaction costs of institutional investors; (2) Algorithmic trading can both decrease market volatility by reducing the impact of large orders, and improve market liquidity by generating realtime updating limit orders; (3) The growing message traffic caused by algorithmic trading will not exceed the system capacity of Shanghai Stock Exchange in the prometaphase. (Joint with Yucaho Wang)
Speaker: Nicolo Pecora
Title: Boom and burst in housing market with heterogeneous agents
Abstract: We present a stylized house market using a partial equilibrium model in which the rational expectation hypothesis is relaxed in favor of an AgentBased approach. The chartistfundamentalist mechanism enhances the behavioral foundation of the expectations and it contributes to replicate the recent house price dynamics. The beliefs dynamics propagate economic shocks and contribute to replicate the empirical evidence belief dynamic can drive house prices away from fundamentals, so that low interest rates can give strength to a house price boom. Moreover, we also analyze the role of interest rate via a 'pseudo' Taylor rule and, according to the model, the US price boom could have largely been avoided if real interest rates had decreased by less, reducing the volatility and the distortion in house price. Even with such a simple model, we can show that the behavioral foundation of the house market has important influence on the macroeconomy.
Speaker: Arturo Ormeno
Title: Using Survey Data on In
ation Expectations in the Estimation of Learning and Rational Expectations Models
Abstract: Do survey data on in
ation expectations contain useful information for estimating macroeconomic models? I address this question by using survey data in the New Keynesian model by Smets and Wouters (2007) to estimate and compare its performance when solved under the assumptions of Rational Expectations and learning. This information serves as an additional moment restriction and helps to determine the forecasting model for inflation that agents use under learning. My results reveal that the predictive power of this model is improved when using both survey data and an admissible learning rule for the formation of inflation expectations.
Speaker: Michiel van de Leur
Title: Meanreverting response to shocks in financial markets
Abstract: The commonly used theory on stock price processes includes strict assumptions. One of these assumption is that past behaviour of stocks has no influence on the future behaviour, i.e. it is impossible to predict future stock returns. To test whether this indeed holds, we investigate the increments $\delta S\beta\delta I$ after moments with large movements and try to predict the sign of the postshock increment with more than $50\%$ certainty. By subtracting $\beta\delta I$ the market risk is removed from the stock return. This could give evidence to reject the assumption of independent stock increments and moreover it might be profitable to capitalize on this predictability.
Based on our findings we change the drift of the commonly used geometric Brownian Motion model for the stock price process in such a way that it differs in postshock time intervals. In this model shocks of the above process are partly reversed in the following 20 minutes. Therefore the assumption of independent increments is rejected and the behaviour after a shock depends on the sign and the strength of the shock.
Speaker: David Kopanyi
Title: Heterogeneous Learning in a Bertrand Oligopoly
Abstract: We analyze the interaction between least squares learning and gradient learning in a Bertrand oligopoly with differentiated goods. Firms have incomplete information about the demand specification. They use one of the learning methods for deciding which price to ask. We analyze four different setups. In a pure OLSlearning oligopoly firms always converge to a \emph{selfsustaining equilibrium} in which their expected and actual demands coincide at the prices they charge. The SSEs constitute a much larger set than the Nash equilibrium. When every firm applies gradient learning and the speed of adjustment is sufficiently low, then firms reach the Nash equilibrium. For gradient learning with a high speed of adjustment, however, we observe highperiod cycles or quasiperiodic dynamics. In a heterogeneous setting where some firms use OLS learning while others apply gradient learning, the dynamical properties of the model depend on the distribution of the learning methods among firms: an increase in the number of gradient learners may destabilize the system. The stable gradient learning disciplines OLS learning: OLS prices are more concentrated around the Nash equilibrium price compared to the case of a pure OLSlearning oligopoly. When firms are allowed to switch between the learning methods, a cyclical switching between the learning methods may be observed when the stability of gradient learning changes as the number of gradient learners varies. As firms switch from the unstable gradient learning to OLS learning, gradient learning becomes stable, yielding a higher average profit. When firms start switching back to gradient learning the method becomes unstable again and the process repeats itself.
Speaker: Benjamin Kemper
Title: Utilitybased Appointment Scheduling
Abstract: When setting up an appointment schedule, one aims at achieving a proper balance between the agents' interests: if the system is frequently idle, then it is not functioning in a costeffective manner for the service provider, whereas if it is virtually always busy, the customers' waiting times may become substantial. We investigate schemes that align these `disutilities' (or, losses) in such a way that they are sequentially (i.e., on a percustomer basis) minimized. The approach carries over to a broad class of loss functions and any service time distribution; it is even allowed that the customers' service times have different distributions.
Next, we develop a criterion that yields the optimal order in case the service time distributions belong to a scale family, such as the exponential family. We prove that customers should then be scheduled in nondecreasing order of their scale parameter. Our findings are illustrated by a number of numerical examples.
&
Speaker: Marit Schoonhoven
Title: A Robust Standard Deviation Control Chart
Abstract: A standard deviation control chart is applied in practice to monitor the standard deviation of a process characteristic. It is a graph of estimates of the process standard deviation on the vertical axis plotted against time on the horizontal axis, supplemented with an upper and a lower control limit. When an estimate falls outside the control limits, it is likely that the standard deviation of the process has shifted. In order to determine the control limits, it is necessary to estimate the process standard deviation based on data taken when the process is in control.
In our study we evaluate the effect of the standard deviation estimator used to determine the control limits. Various standard deviation estimators are used for the control limits, and the effect on control chart performance is assessed. We include the uncontaminated case, i.e. the estimation data are in control, as well as several types of disturbances. Finally, we propose an algorithm to estimate the standard deviation that is intuitive in its use and performs well against various types of disturbances.
Seminar cancelled
Speaker: Marcin Wolski
Title: The influence of systemic risk on sovereign stability: Evidence from the Eurozone
Abstract: The financial crisis 2007/09 shed a new light on the importance of Systemic Risk (SR) (Schwarcz, 2008). In short, SR can be recognised as a risk of disruption to financial services that results from an impairment of the financial system (BIS, 2011). This Thesis quantifies SR in the Eurozone in years 20062010 using the methodology proposed by Acharya et al. (2010). The results prove that over this period SR was fluctuating. On the basis of that three sub periods were distinguished: the uncertainty period, between first quarter 2006 and second quarter 2007, the crisis period, until second quarter 2009 and the calm period, until the end of the time span. This classification is in line with the literature (Getter et al., 2007).
The influence of SR on sovereign stability have been measured using the system GMM estimator. The results show that SR was positively and significantly associated with the longer term Sovereign Credit Default Swaps (SCDS). Shorter term SCDS are also positively affected by SR, however, its coefficients are not significant. This was mainly caused by the vulnerability to short term shocks, bail out programs and ”humped” yield curve observed in the dataset. Moreover, the analysis shows that the effect of SR on SCDS changes when a certain threshold level is being reached.
Speaker: Chengyao Wu
Title: Information Cost, Memory Length and Market Instability
Abstract: In this paper we study the instability of a stock market with a modi ed version of Diks and Dindo (2008)'s model where the market is characterized by nonlinear interactions between informed traders and uninformed traders. In the interaction of heterogeneous agents, we substitute the logisticstrategy switching by replicator dynamics, and this modi cation makes the model more suitable for describing deviations from fundamental prices, as well as more robust. One goal of our paper is to use this model to explore if the arrival of new information (news) and investor behavior have an effect on the market instability. A second, related, goal is to study the way markets absorb new information, especially when the market is unstable and the price is far from being fully informative. We found that the dynamics become locally unstable and prices may deviate far from the fundamental price, routing to chaos through bifurcation, with increasing in the information cost or decreasing memory length of the uninformed traders. On the inuence of the arrival of news, the more expensive the new information is, the more effect news has on the market.
Speaker: Paolo Zeppini
Title: Competing Technologies: A Discrete Choice Model
Abstract: We propose a discrete choice model of technology competition with network externalities and social interactions, that is able to reproduce stylized facts of technology competition, like path dependence and lockin, as emergent phenomena. An application to environmental policy describes the competition of "dirty" and "clean" technologies. Social interactions strongly a ect the outcome of an environmental policy aimed at reducing pollution. Unexpectedly, a tougher policy may fail in this attempt, leading to cyclical dynamics. The interaction of technology competition and endogenous technological progress leads to decision externalities strongly a ecting the overall rate of technological progress. The full model brings together environmental policy and technological progress. We nd that sustained progress of the clean technology is needed beside an environmental policy in order to unlock the market from the dirty technology. Moreover, lower social interactions and network externalities allow for a less stringent policy. This fact suggests a multilevel approach to environmental policy, that also addresses technological standards and infrastructures.
Speaker: Tomasz A. Makarewicz
Title: Evolution of Learning Mechanisms in the Ultimatum Game
Abstract: In this paper, I discuss the relevance of learning models for ultimatumgame experiments, in which subjects tend to play offequilibrium strategies. I consider two benchmark learning mechanisms, reinforcement learning which is a possible interpretation for experimental results; and fictitious learning which is well known for its efficient use of hypothetical reasoning. I design a model in which these two learning algorithms compete within a population through evolutionary replicator dynamics. Analytical investigation based on miniultimatum game shows that both mechanisms are likely to converge to the same behavior; and that the subgame perfect Nash equilibrium (SPNE) is a likely (but not a unique) fixed point. Further simulations for finite populations indicate that the slow convergence of reinforcement learning makes it perform poorly in the shortrun evolutionary competition with fictitious learning. Regardless of the population size, the system is attracted to the SPNE, which therefore proves to be a robust outcome in the heterogeneous environment of learning mechanisms. This negative result indicates that the learningbased argument, applied directly to the ultimatum game, may be insufficient to explain human behavior in experiments.
2017
 Bao, T., Hommes, C. H., & Makarewicz, T. A. (2017). Bubble formation and (in)efficient markets in learningtoforecast and optimize experiments. Economic Journal. [details]
2012
 Diks, C., & Makarewicz, T. (2012). Initial predictions in learningtoforecast experiment. In A. Teglio, S. Alfarano, E. CamachoCuena, & M. GinésVilar (Eds.), Managing market complexity: the approach of artificial economics (pp. 223235). (Lecture notes in economics and mathematical systems; No. 662). Berlin / Heidelberg: SpringerVerlag. DOI: 10.1007/9783642313011_18 [details]
2014
 Makarewicz, T. A. (2014). Learning to forecast: Genetic algorithms and experiments Amsterdam: Tinbergen Institute [details / files]
2015
 Anufriev, M., Hommes, C., & Makarewicz, T. (2015). Simple forecasting heuristics that make us smart: evidence from different market experiments. (CeNDEF working paper; No. 1507). Amsterdam: CeNDEF, University of Amsterdam. [details] [PDF]
 Hommes, C., Makarewicz, T., Massaro, D., & Smits, T. (2015). Genetic Algorithm Learning in a New Keynesian Macroeconomic Setup. (CeNDEF Working paper; No. 1501). Amsterdam: University of Amsterdam. [details]
 Makarewicz, T. (2015). Networks of Heterogeneous Expectations in an Asset Pricing Market. (CeNDEF working paper; No. 1508). Amsterdam: CeNDEF, University of Amsterdam. [details] [PDF]
2014
 Bao, T., Hommes, C., & Makarewicz, T. (2014). Bubble formation and (in)efficient markets in learningtoforecast and optimize experiments. (CeNDEF Working Paper; No. 1401). Amsterdam: Universiteit van Amsterdam. [details]