Inference in differences-in-differences with few treated groups and heteroskedasticity

Differences-in-Differences (DID) is one of the most widely used identification strategies in applied economics. However, how to draw inferences in DID models when there are few treated groups remains an open question. We show that the usual inference methods used in DID models might not perform well when there are few treated groups and errors are heteroskedastic. In particular, we show that when there is variation in the number of observations per group, inference methods designed to work when there are few treated groups tend to (under-) over-reject the null hypothesis when the treated groups are (large) small relative to the control groups. This happens because larger groups tend to have lower variance, generating heteroskedasticity in the group x time aggregate DID model. We provide evidence from Monte Carlo simulations and from placebo DID regressions with the American Community Survey (ACS) and the Current Population Survey (CPS) datasets to show that this problem is relevant even in datasets with large numbers of observations per group. We then derive an alternative inference method that provides accurate hypothesis testing in situations where there are few treated groups (or even just one) and many control groups in the presence of heteroskedasticity. Our method assumes that we can model the heteroskedasticity of a linear combination of the errors. We show that this assumption can be satisfied without imposing strong assumptions on the errors in common DID applications. With many pre-treatment periods, we show that this assumption can be relaxed. Instead, we provide an alternative inference method that relies on strict stationarity and ergodicity of the time series. Finally, we consider two recent alternatives to DID when there are many pre-treatment periods. We extend our inference methods to linear factor models when there are few treated groups. We also derive conditions under which a permutation test for the synthetic control estimator proposed by Abadie et al. (2010) is robust to heteroskedasticity and propose a modification on the test statistic that provided a better heteroskedasticity correction in our simulations.

Inference in differences-in-differences with few treated groups and heteroskedasticity

Differences-in-Differences (DID) is one of the most widely used identification strategies in applied economics. However, how to draw inferences in DID models when there are few treated groups remains an open question. We show that the usual inference methods used in DID models might not perform well when there are few treated groups and errors are heteroskedastic. In particular, we show that when there is variation in the number of observations per group, inference methods designed to work when there are few treated groups tend to (under-) over-reject the null hypothesis when the treated groups are (large) small relative to the control groups. This happens because larger groups tend to have lower variance, generating heteroskedasticity in the group x time aggregate DID model. We provide evidence from Monte Carlo simulations and from placebo DID regressions with the American Community Survey (ACS) and the Current Population Survey (CPS) datasets to show that this problem is relevant even in datasets with large numbers of observations per group. We then derive an alternative inference method that provides accurate hypothesis testing in situations where there are few treated groups (or even just one) and many control groups in the presence of heteroskedasticity. Our method assumes that we know how the heteroskedasticity is generated, which is the case when it is generated by variation in the number of observations per group. With many pre-treatment periods, we show that this assumption can be relaxed. Instead, we provide an alternative application of our method that relies on assumptions about stationarity and convergence of the moments of the time series. Finally, we consider two recent alternatives to DID when there are many pre-treatment groups. We extend our inference method to linear factor models when there are few treated groups. We also propose a permutation test for the synthetic control estimator that provided a better heteroskedasticity correction in our simulations than the test suggested by Abadie et al. (2010).

The costs of education, longevity and the poverty of nations

This paper explores the distortions on the cost of education, associated with government policies and institutional factors, as an additional determinant of cross-country income differences. Agents are finitely lived and the model takes into account life-cycle features of human capital accumulation. There are two sectors, one producing goods and the other providing educational services. The model is calibrated and simulated for 89 economies. We find that human capital taxation has a relevant impact on incomes, which is amplified by its indirect effect on returns to physical capital. Life expectancy plays an important role in determining long-run output: the expansion of the population working life increases the present value of the flow of wages, which induces further human capital investment and raises incomes. Although in our simulations the largest gains are observed when productivity is equated across countries, changes in longevity and in the incentives to educational investment are too relevant to ignore.

Otimização de alavancagem e gestão de risco em estratégias long-short

Alavancagem em hedge funds tem preocupado investidores e estudiosos nos últimos anos. Exemplos recentes de estratégias desse tipo se mostraram vantajosos em períodos de pouca incerteza na economia, porém desastrosos em épocas de crise. No campo das finanças quantitativas, tem-se procurado encontrar o nível de alavancagem que otimize o retorno de um investimento dado o risco que se corre. Na literatura, os estudos têm se mostrado mais qualitativos do que quantitativos e pouco se tem usado de métodos computacionais para encontrar uma solução. Uma forma de avaliar se alguma estratégia de alavancagem aufere ganhos superiores do que outra é definir uma função objetivo que relacione risco e retorno para cada estratégia, encontrar as restrições do problema e resolvê-lo numericamente por meio de simulações de Monte Carlo. A presente dissertação adotou esta abordagem para tratar o investimento em uma estratégia long-short em um fundo de investimento de ações em diferentes cenários: diferentes formas de alavancagem, dinâmicas de preço das ações e níveis de correlação entre esses preços. Foram feitas simulações da dinâmica do capital investido em função das mudanças dos preços das ações ao longo do tempo. Considerou-se alguns critérios de garantia de crédito, assim como a possibilidade de compra e venda de ações durante o período de investimento e o perfil de risco do investidor. Finalmente, estudou-se a distribuição do retorno do investimento para diferentes níveis de alavancagem e foi possível quantificar qual desses níveis é mais vantajoso para a estratégia de investimento dadas as restrições de risco.

Eficiência da magic formula de value investing no mercado brasileiro

O objetivo deste trabalho é realizar procedimento de back-test da Magic Formula na Bovespa, reunindo evidências sobre violações da Hipótese do Mercado Eficiente no mercado brasileiro. Desenvolvida por Joel Greenblatt, a Magic Formula é uma metodologia de formação de carteiras que consiste em escolher ações com altos ROICs e Earnings Yields, seguindo a filosofia de Value Investing. Diversas carteiras foram montadas no período de dezembro de 2002 a maio de 2014 utilizando diferentes combinações de número de ativos por carteira e períodos de permanência. Todas as carteiras, independentemente do número de ativos ou período de permanência, apresentaram retornos superiores ao Ibovespa. As diferenças entre os CAGRs das carteiras e o do Ibovespa foram significativas, sendo que a carteira com pior desempenho apresentou CAGR de 27,7% contra 14,1% do Ibovespa. As carteiras também obtiveram resultados positivos após serem ajustadas pelo risco. A pior razão retorno-volatilidade foi de 1,2, comparado a 0,6 do Ibovespa. As carteiras com pior pontuação também apresentaram bons resultados na maioria dos cenários, contrariando as expectativas iniciais e os resultados observados em outros trabalhos. Adicionalmente foram realizadas simulações para diversos períodos de 5 anos com objetivo de analisar a robustez dos resultados. Todas as carteiras apresentaram CAGR maior que o do Ibovespa em todos os períodos simulados, independentemente do número de ativos incluídos ou dos períodos de permanência. Estes resultados indicam ser possível alcançar retornos acima do mercado no Brasil utilizando apenas dados públicos históricos. Esta é uma violação da forma fraca da Hipótese do Mercado Eficiente.

Cherry picking with synthetic controls

The synthetic control (SC) method has been recently proposed as an alternative method to estimate treatment e ects in comparative case studies. Abadie et al. [2010] and Abadie et al. [2015] argue that one of the advantages of the SC method is that it imposes a data-driven process to select the comparison units, providing more transparency and less discretionary power to the researcher. However, an important limitation of the SC method is that it does not provide clear guidance on the choice of predictor variables used to estimate the SC weights. We show that such lack of speci c guidances provides signi cant opportunities for the researcher to search for speci cations with statistically signi cant results, undermining one of the main advantages of the method. Considering six alternative speci cations commonly used in SC applications, we calculate in Monte Carlo simulations the probability of nding a statistically signi cant result at 5% in at least one speci cation. We nd that this probability can be as high as 13% (23% for a 10% signi cance test) when there are 12 pre-intervention periods and decay slowly with the number of pre-intervention periods. With 230 pre-intervention periods, this probability is still around 10% (18% for a 10% signi cance test). We show that the speci cation that uses the average pre-treatment outcome values to estimate the weights performed particularly bad in our simulations. However, the speci cation-searching problem remains relevant even when we do not consider this speci cation. We also show that this speci cation-searching problem is relevant in simulations with real datasets looking at placebo interventions in the Current Population Survey (CPS). In order to mitigate this problem, we propose a criterion to select among SC di erent speci cations based on the prediction error of each speci cations in placebo estimations

Business competition dynamics: agent-based modeling simulations of firms in search of economic performance

A intenção deste trabalho é explorar dinâmicas de competição por meio de “simulação baseada em agentes”. Apoiando-se em um crescente número de estudos no campo da estratégia e teoria das organizações que utilizam métodos de simulação, desenvolveu-se um modelo computacional para simular situações de competição entre empresas e observar a eficiência relativa dos métodos de busca de melhoria de desempenho teorizados. O estudo também explora possíveis explicações para a persistência de desempenho superior ou inferior das empresas, associados às condições de vantagem ou desvantagem competitiva