Archives

  • 2018-07
  • 2018-10
  • 2018-11
  • 2019-04
  • 2019-05
  • 2019-06
  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2019-12
  • 2020-01
  • 2020-02
  • 2020-03
  • 2020-04
  • 2020-05
  • 2020-06
  • 2020-07
  • 2020-08
  • 2020-09
  • 2020-10
  • 2020-11
  • 2020-12
  • 2021-01
  • 2021-02
  • 2021-03
  • 2021-04
  • 2021-05
  • 2021-06
  • 2021-07
  • 2021-08
  • 2021-09
  • 2021-10
  • 2021-11
  • 2021-12
  • 2022-01
  • 2022-02
  • 2022-03
  • 2022-04
  • 2022-05
  • 2022-06
  • 2022-07
  • 2022-08
  • 2022-09
  • 2022-10
  • 2022-11
  • 2022-12
  • 2023-01
  • 2023-02
  • 2023-03
  • 2023-04
  • 2023-05
  • 2023-06
  • 2023-08
  • 2023-09
  • 2023-10
  • 2023-11
  • 2023-12
  • 2024-01
  • 2024-02
  • 2024-03
  • 2024-04
  • 2024-05
  • 2024-06
  • 2024-07
  • 2024-08
  • 2024-09
  • 2024-10

    • In recent years studies in finance and

    2018-10-30

    In recent years studies in finance and in econometrics have tried to produce accurate forecasts of the term structure. Duffee (2002) uses the dynamic properties of affine models and concludes that his forecasts are dominated by the random walk model. The forecast underperformance of affine models stimulated Diebold and Li (2006) to adapt the static Nelson and Siegel (1987) factor model into a dynamic variant. Their results show that the model can beat the random walk, with the exception of the 1-month ahead horizon. After Diebold and Li (2006), many authors have developed different dynamic approaches of Nelson–Siegel model and have reported improvements in forecasting. Pooter (2007) adds new factors. Koopman et al. (2007) introduce time-varying parameters. Diebold et al. (2008) extend the model to a global context, modeling a large set of country yield curves in a framework that allows for both global and country-specific factors. The interaction between macroeconomics and the yield curve has also generated promising results. Prominent examples are Evans and Marshall (1998), Ang and Piazzesi (2003), Hordahl et al. (2006), Diebold and Rudebush (2006), Moench (2008), Ludvingson and Ng (2009) and Pooter et al. (2010). It is important to point out that the use of macroeconomic information is beyond the scope of this study. Despite good statistical properties, the models based on Nelson–Siegel framework do not impose the arbitrage-free restriction, which is a fundamental concept in the literature of bond pricing analysis. In order to solve this problem, Christensen et al. (2010) have developed an affine model that maintain the factor structure of Nelson and Siegel (1987), the so-called Arbitrage-Free Dynamic Nelson–Siegel. They conclude that improvements in predictive performance are achieved from the imposition of the absence of arbitrage. However, Duffee (2011) argues that this restriction is irrelevant for forecast gains and it ARCA only provides a mechanism to specify a functional form for risk compensation. Bowsher and Meeks (2008) do not adopt the Nelson–Siegel framework. Instead, they introduce a factor model, known as Functional Signal Plus Noise with an Equilibrium Correction Model. Splines are used to model the yield curve. Besides, they argue that the knots of these splines act as factors. The results indicate that this model outperforms the random walk mainly at short-term horizons. The concept of forecast combination was introduced in Bates and Granger (1969). They showed that combinations of individual forecasts often outperform individual forecasts. One possible reason is that the combination of forecasts adds value in the presence of structural breaks (Diebold and Pauly, 1987; Hendry and Clements, 2004). A number of papers (Diebold and Lopez, 1996; Stock and Watson, 2004; Timmermann, 2006) also defends that individual forecast models may be subject to misspecification biases of unknown form. The remainder of the paper is structured as follows. Section 2 presents the forecast models used in this paper. Section 3 describes the alternative combining methods. Section 4 introduces the data. Section 5 presents the forecast procedures and the evaluating criteria. Section 6 shows the main results of this paper. It evaluates the performance of individual models and combining methods. Finally, Section 7 concludes the paper.
    Forecast combining methods In order to test the predictive capacity of the models, we split the data into in-sample (the first R observations) and out-of-sample (the last P observations). Furthermore, in some schemes, namely inverted MSFE, WLS and combining via MCS, we need an extra out-of-sample period (training period) for computing the weights of the individual forecast models. In this case, we use the first P0 out-of-sample observations. More specifically, we calculate the weights using the individual out-of-sample forecasts, , and the observations, , both available from the beginning of the training out-of-sample period to time t. In turn, we use the P1 post-training out-of-sample observations to compare the forecast performance of the individual models and combining schemes.