| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.36 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O objetivo deste estudo é averiguar se a previsão out-of-sample do retorno do
portfólio composto por 27 futuros de commodities, construído por Asness et al.
(2013), pode ser aperfeiçoada usando técnicas de previsão no domínio da
frequência. Além das variáveis preditivas propostas por Lutzenberger (2014),
que compreendem características de ações, de obrigações/títulos,
macroeconómicas e de commodities, seguimos o procedimento de Faria and
Verona (2018b) e consideramos também as suas frequências individuais como
potenciais variáveis preditivas. Para tal, extraímos as componentes de frequência
das variáveis preditivas usando métodos de filtragem de wavelets e estudamos o
seu desempenho na previsão in-sample e out-of-sample do portefólio.
Os resultados obtidos sugerem que o default return spread e o prémio de capital
acumulado têm empiricamente um poder robusto na previsão out-of-sample do
portfólio de futuros de commodities para um horizonte temporal de um mês. O
default return spread verifica uma performance superior à das outras variáveis
preditivas assim que a relação entre a sua frequência (de ciclo de negócio) e o
portfólio de commodities é tida em consideração. Similarmente, assim que a
relação entre a (baixa) frequência do prémio de capital acumulado e o portfólio
de commodities é considerada, este verifica uma melhor performance do que as
outras variáveis. Estas variáveis preditivas são também estatisticamente
significantes na previsão in-sample. A componente de baixa frequência do default
return spread e a componente de alta frequência da volatilidade de commodities
são estatisticamente significantes na previsão in-sample e out-of-sample para o
mesmo horizonte temporal.
The aim of this study is to test whether the out-of-sample predictability of the return on the equal-weighted portfolio of 27 commodity futures constructed by Asness et al. (2013) can be improved by using frequency domain forecasting techniques. We attempt to identify variables that show predictive power over these returns. Besides the candidate predictors proposed by Lutzenberger (2014), which comprise of stock, bond, macroeconomic, and commodity features, we follow the approach of Faria and Verona (2018b) and consider their individual frequency components as potential predictors. To do so, we extract frequency components of the predictors using wavelet filtering methods and study both the in-sample fit and out-of-sample forecasting performance of the portfolio. Our results suggest that the default return spread and the cumulative equity premium have a remarkably robust empirical OOS forecasting power of the commodity futures portfolio over a forecasting horizon of one month. They fit the data better in an out-of-sample forecasts exercise than the other predictors, once the (business cycle) frequency relationship between default return spread and commodity portfolio is taken into account and once the (low) frequency relationship between cumulative equity premium and commodity portfolio is taken into account. They also present in-sample statistically significance. Additionally, we find that the low frequency component of the default return spread and the high frequency component of commodity volatility are good predictors of our portfolio both in-sample and out-of-sample over the same forecasting horizon.
The aim of this study is to test whether the out-of-sample predictability of the return on the equal-weighted portfolio of 27 commodity futures constructed by Asness et al. (2013) can be improved by using frequency domain forecasting techniques. We attempt to identify variables that show predictive power over these returns. Besides the candidate predictors proposed by Lutzenberger (2014), which comprise of stock, bond, macroeconomic, and commodity features, we follow the approach of Faria and Verona (2018b) and consider their individual frequency components as potential predictors. To do so, we extract frequency components of the predictors using wavelet filtering methods and study both the in-sample fit and out-of-sample forecasting performance of the portfolio. Our results suggest that the default return spread and the cumulative equity premium have a remarkably robust empirical OOS forecasting power of the commodity futures portfolio over a forecasting horizon of one month. They fit the data better in an out-of-sample forecasts exercise than the other predictors, once the (business cycle) frequency relationship between default return spread and commodity portfolio is taken into account and once the (low) frequency relationship between cumulative equity premium and commodity portfolio is taken into account. They also present in-sample statistically significance. Additionally, we find that the low frequency component of the default return spread and the high frequency component of commodity volatility are good predictors of our portfolio both in-sample and out-of-sample over the same forecasting horizon.
Description
Keywords
Previsão Retorno de ativos Commodities Domínio da frequência Wavelet Predictability Asset returns Frequency domain