Evolutionary Testing of Autonomous Software Agents

A system built in terms of autonomous agents may require even greater correctness assurance than one which is merely reacting to the immediate control of its users. Agents make substantial decisions for themselves, so thorough testing is an important consideration. However, autonomy also makes testing harder; by their nature, autonomous agents may react in different ways to the same inputs over time, because, for instance they have changeable goals and knowledge. For this reason, we argue that testing of autonomous agents requires a procedure that caters for a wide range of test case contexts, and that can search for the most demanding of these test cases, even when they are not apparent to the agents’ developers. In this paper, we address this problem, introducing and evaluating an approach to testing autonomous agents that uses evolutionary optimization to generate demanding test cases.

Extension And Testing Of A 2D Hydrodynamic Model For Direct Rainfall Runoff Simulation

While the simulation of flood risks originating from the overtopping of river banks is well covered within continuously evaluated programs to improve flood protection measures, flash flooding is not. Flash floods are triggered by short, local thunderstorm cells with high precipitation intensities. Small catchments have short response times and flow paths and convective thunder cells may result in potential flooding of endangered settlements. Assessing local flooding and pathways of flood requires a detailed hydraulic simulation of the surface runoff. Hydrological models usually do not incorporate surface runoff at this detailedness but rather empirical equations are applied for runoff detention. In return 2D hydrodynamic models usually do not allow distributed rainfall as input nor are any types of soil/surface interaction implemented as in hydrological models. Considering several cases of local flash flooding during the last years the issue emerged for practical reasons but as well as research topics to closing the model gap between distributed rainfall and distributed runoff formation. Therefore, a 2D hydrodynamic model, depth-averaged flow equations using the finite volume discretization, was extended to accept direct rainfall enabling to simulate the associated runoff formation. The model itself is used as numerical engine, rainfall is introduced via the modification of waterlevels at fixed time intervals. The paper not only deals with the general application of the software, but intends to test the numerical stability and reliability of simulation results. The performed tests are made using different artificial as well as measured rainfall series as input. Key parameters of the simulation such as losses, roughness or time intervals for water level manipulations are tested regarding their impact on the stability.

Testing the externalities hypothesis of endogenous growth using cointegration

The initial endogenous growth models emphasized the importance of externaI effects in explaining sustainable growth across time. Empirically, this hypothesis can be confirmed if the coefficient of physical capital per hour is unity in the aggregate production function. Although cross-section results concur with theory, previous estimates using time series data rejected this hypothesis, showing a small coefficient far from unity. It seems that the problem lies not with the theory but with the techniques employed, which are unable to capture low frequency movements in high frequency data. This paper uses cointegration - a technique designed to capture the existence of long-run relationships in multivariate time series - to test the externalities hypothesis of endogenous growth. The results confirm the theory' and conform to previous cross-section estimates. We show that there is long-run proportionality between output per hour and a measure of capital per hour. U sing this result, we confmn the hypothesis that the implied Solow residual can be explained by government expenditures on infra-structure, which suggests a supply side role for government affecting productivity and a decrease on the extent that the Solow residual explains the variation of output.

Testing unit root based on partially adaptive estimation

This paper proposes unit tests based on partially adaptive estimation. The proposed tests provide an intermediate class of inference procedures that are more efficient than the traditional OLS-based methods and simpler than unit root tests based on fully adptive estimation using nonparametric methods. The limiting distribution of the proposed test is a combination of standard normal and the traditional Dickey-Fuller (DF) distribution, including the traditional ADF test as a special case when using Gaussian density. Taking into a account the well documented characteristic of heavy-tail behavior in economic and financial data, we consider unit root tests coupled with a class of partially adaptive M-estimators based on the student-t distributions, wich includes te normal distribution as a limiting case. Monte Carlo Experiments indicate that, in the presence of heavy tail distributions or innovations that are contaminated by outliers, the proposed test is more powerful than the traditional ADF test. We apply the proposed test to several macroeconomic time series that have heavy-tailed distributions. The unit root hypothesis is rejected in U.S. real GNP, supporting the literature of transitory shocks in output. However, evidence against unit roots is not found in real exchange rate and nominal interest rate even haevy-tail is taken into a account.

Testing production functions used in empirical growth studies

We estimate and test two alternative functional forms, which have been used in the growth literature, representing the aggregate production function for a panel of countries: the model of Mankiw, Romer and Weil (Quarterly Journal of Economics, 1992), and a mincerian formulation of schooling-returns to skills. Estimation is performed using instrumental-variable techniques, and both functional forms are confronted using a Box-Cox test, since human capital inputs enter in levels in the mincerian specification and in logs in the extended neoclassical growth model.

Testing covariance stationarity

In this paper, we show that the widely used stationarity tests such as the KPSS test have power close to size in the presence of time-varying unconditional variance. We propose a new test as a complement of the existing tests. Monte Carlo experiments show that the proposed test possesses the following characteristics: (i) In the presence of unit root or a structural change in the mean, the proposed test is as powerful as the KPSS and other tests; (ii) In the presence a changing variance, the traditional tests perform badly whereas the proposed test has high power comparing to the existing tests; (iii) The proposed test has the same size as traditional stationarity tests under the null hypothesis of stationarity. An application to daily observations of return on US Dollar/Euro exchange rate reveals the existence of instability in the unconditional variance when the entire sample is considered, but stability is found in subsamples.

Testing the hypothesis of contagion using multivariate volatility models

The aim of this paper is to test whether or not there was evidence of contagion across the various financial crises that assailed some countries in the 1990s. Data on sovereign debt bonds for Brazil, Mexico, Russia and Argentina were used to implement the test. The contagion hypothesis is tested using multivariate volatility models. If there is any evidence of structural break in volatility that can be linked to financial crises, the contagion hypothesis will be confirmed. Results suggest that there is evidence in favor of the contagion hypothesis.

Testing the long-run implications of the expectation hypothesis using cointegration techniques with structural change

This paper investigates whether or not multivariate cointegrated process with structural change can describe the Brazilian term structure of interest rate data from 1995 to 2006. In this work the break point and the number of cointegrated vector are assumed to be known. The estimated model has four regimes. Only three of them are statistically different. The first starts at the beginning of the sample and goes until September of 1997. The second starts at October of 1997 until December of 1998. The third starts at January of 1999 and goes until the end of the sample. It is used monthly data. Models that allows for some similarities across the regimes are also estimated and tested. The models are estimated using the Generalized Reduced-Rank Regressions developed by Hansen (2003). All imposed restrictions can be tested using likelihood ratio test with standard asymptotic 1 qui-squared distribution. The results of the paper show evidence in favor of the long run implications of the expectation hypothesis for Brazil.

Testing the optimality of aggregate consumption decisions: is there rule-of-thumb behavior?

Consumption is an important macroeconomic aggregate, being about 70% of GNP. Finding sub-optimal behavior in consumption decisions casts a serious doubt on whether optimizing behavior is applicable on an economy-wide scale, which, in turn, challenge whether it is applicable at all. This paper has several contributions to the literature on consumption optimality. First, we provide a new result on the basic rule-of-thumb regression, showing that it is observational equivalent to the one obtained in a well known optimizing real-business-cycle model. Second, for rule-of-thumb tests based on the Asset-Pricing Equation, we show that the omission of the higher-order term in the log-linear approximation yields inconsistent estimates when lagged observables are used as instruments. However, these are exactly the instruments that have been traditionally used in this literature. Third, we show that nonlinear estimation of a system of N Asset-Pricing Equations can be done efficiently even if the number of asset returns (N) is high vis-a-vis the number of time-series observations (T). We argue that efficiency can be restored by aggregating returns into a single measure that fully captures intertemporal substitution. Indeed, we show that there is no reason why return aggregation cannot be performed in the nonlinear setting of the Pricing Equation, since the latter is a linear function of individual returns. This forms the basis of a new test of rule-of-thumb behavior, which can be viewed as testing for the importance of rule-of-thumb consumers when the optimizing agent holds an equally-weighted portfolio or a weighted portfolio of traded assets. Using our setup, we find no signs of either rule-of-thumb behavior for U.S. consumers or of habit-formation in consumption decisions in econometric tests. Indeed, we show that the simple representative agent model with a CRRA utility is able to explain the time series data on consumption and aggregate returns. There, the intertemporal discount factor is significant and ranges from 0.956 to 0.969 while the relative risk-aversion coefficient is precisely estimated ranging from 0.829 to 1.126. There is no evidence of rejection in over-identifying-restriction tests.

Testing production functions used in empirical growth studies

We estimate and test two alternative functional forms representing the aggregate production function for a panel of countries: the extended neoclassical growth model, and a mincerian formulation of schooling-returns to skills. Estimation is performed using instrumentalvariable techniques, and both functional forms are confronted using a Box-Cox test, since human capital inputs enter in levels in the mincerian specification and in logs in the extended neoclassical growth model. Our evidence rejects the extended neoclassical growth model in favor of the mincerian specification, with an estimated capital share of about 42%, a marginal return to education of about 7.5% per year, and an estimated productivity growth of about 1.4% per year. Differences in productivity cannot be disregarded as an explanation of why output per worker varies so much across countries: a variance decomposition exercise shows that productivity alone explains 54% of the variation in output per worker across countries.

Testing the Markov property with ultra high frequency financial data

This paper develops a framework to test whether discrete-valued irregularly-spaced financial transactions data follow a subordinated Markov process. For that purpose, we consider a specific optional sampling in which a continuous-time Markov process is observed only when it crosses some discrete level. This framework is convenient for it accommodates not only the irregular spacing of transactions data, but also price discreteness. Further, it turns out that, under such an observation rule, the current price duration is independent of previous price durations given the current price realization. A simple nonparametric test then follows by examining whether this conditional independence property holds. Finally, we investigate whether or not bid-ask spreads follow Markov processes using transactions data from the New York Stock Exchange. The motivation lies on the fact that asymmetric information models of market microstructures predict that the Markov property does not hold for the bid-ask spread. The results are mixed in the sense that the Markov assumption is rejected for three out of the five stocks we have analyzed.

A Common-feature approach for testing present-value restrictions with financial data

It is well known that cointegration between the level of two variables (labeled Yt and yt in this paper) is a necessary condition to assess the empirical validity of a present-value model (PV and PVM, respectively, hereafter) linking them. The work on cointegration has been so prevalent that it is often overlooked that another necessary condition for the PVM to hold is that the forecast error entailed by the model is orthogonal to the past. The basis of this result is the use of rational expectations in forecasting future values of variables in the PVM. If this condition fails, the present-value equation will not be valid, since it will contain an additional term capturing the (non-zero) conditional expected value of future error terms. Our article has a few novel contributions, but two stand out. First, in testing for PVMs, we advise to split the restrictions implied by PV relationships into orthogonality conditions (or reduced rank restrictions) before additional tests on the value of parameters. We show that PV relationships entail a weak-form common feature relationship as in Hecq, Palm, and Urbain (2006) and in Athanasopoulos, Guillén, Issler and Vahid (2011) and also a polynomial serial-correlation common feature relationship as in Cubadda and Hecq (2001), which represent restrictions on dynamic models which allow several tests for the existence of PV relationships to be used. Because these relationships occur mostly with nancial data, we propose tests based on generalized method of moment (GMM) estimates, where it is straightforward to propose robust tests in the presence of heteroskedasticity. We also propose a robust Wald test developed to investigate the presence of reduced rank models. Their performance is evaluated in a Monte-Carlo exercise. Second, in the context of asset pricing, we propose applying a permanent-transitory (PT) decomposition based on Beveridge and Nelson (1981), which focus on extracting the long-run component of asset prices, a key concept in modern nancial theory as discussed in Alvarez and Jermann (2005), Hansen and Scheinkman (2009), and Nieuwerburgh, Lustig, Verdelhan (2010). Here again we can exploit the results developed in the common cycle literature to easily extract permament and transitory components under both long and also short-run restrictions. The techniques discussed herein are applied to long span annual data on long- and short-term interest rates and on price and dividend for the U.S. economy. In both applications we do not reject the existence of a common cyclical feature vector linking these two series. Extracting the long-run component shows the usefulness of our approach and highlights the presence of asset-pricing bubbles.