## Changes in Predictive Ability with Mixed Frequency Data.” Working Paper No (2007)

### Cached

### Download Links

Citations: | 8 - 1 self |

### BibTeX

@TECHREPORT{Galvão07changesin,

author = {Ana Beatriz Galvão},

title = {Changes in Predictive Ability with Mixed Frequency Data.” Working Paper No},

institution = {},

year = {2007}

}

### OpenURL

### Abstract

This paper proposes a new regression model — a smooth transition mixed data sampling (STMIDAS) approach — that captures recurrent changes in the ability of a high frequency variable in predicting a variable only available at lower frequency. The model is applied to the use of …nancial variables, such as the slope of the yield curve, the short-rate and stock returns, to forecast US output growth both in- and out-of-sample. I …nd evidence that the use of the predictor sampled weekly improves output growth forecasts, which may also be improved when changes in …nancial variables’predictive power are considered.

### Citations

1072 | A New Approach to The Economic Analysis of Non– Stationary Time Series
- Hamilton
- 1989
(Show Context)
Citation Context ...for longer horizons are then obtained by bootstrap. Another alternative for modelling switching regimes is to make the regimes dependent on a latent variable, which is controlled by a Markov process (=-=Hamilton, 1989-=-). In comparison with this alternative, the STMIDAS has the regime switching to depend on the size and sign of the observable variable. Finally, STMIDAS is able to capture asymmetries in the predictiv... |

724 | Comparing predictive accuracy
- Diebold, Mariano
- 1995
(Show Context)
Citation Context ...ecasters is n 1 P n t=1 ^e2 h;BF;t ^e 2 h;RF;t , where ^e is the forecast error. Using this di¤erential, a t-statistic, computed with a HAC estimator of the variance, is applied for testing the null (=-=Diebold and Mariano, 1995-=-). In the case of non-nested models, such as the case that the RF forecaster has18 no autoregressive term, this statistic has normal distribution. Similar testing is employed for evaluating whether R... |

632 | Asset pricing - Cochrane |

429 |
Non-linear Time Series: A Dynamical System Approach
- Tong
- 1990
(Show Context)
Citation Context ...an and Taylor, 2003). 2.2 Smooth Transition MIDAS Switching regimes are a popular way of modelling nonlinear dynamics in regressions by using piecewise linear regimes linked by a transition function (=-=Tong, 1990-=-). When the transition between regimes is smooth and it depends on the size of an observed transition variable, switchingregime models are called smooth transition regressions (surveyed by Van Dijk, T... |

327 |
Modelling nonlinear economic relationships
- Granger, Teräsvirta
- 1993
(Show Context)
Citation Context ...ability with equation (1) under the null and equation (4) under the alternative is the presence of nuisance parameters under the null, which has a severe impact on the size of usual asymptotic tests (=-=Granger and Teräsvirta, 1993-=-). Fortunately, the testing procedure8 proposed by Luukkonen, Saikkonen and Terasvirta (1988) can be applied. The testing procedure makes use of a Taylor approximation of the logistic function. Using... |

289 | Dividend Yields and Expected Stock Returns: Alternative - Hodrick - 1992 |

195 |
1991):”The term structure as a predictor of real economic activity
- Estrella, Hardouvelis
(Show Context)
Citation Context ... predictors of output growth (Harvey, 1988; Stock and Watson, 2003). Indeed, one of the most popular leading indicators of the US growth is the spread between long-term and short-term interest rates (=-=Estrella and Hardouvelis, 1991-=-; Hamilton and Kim, 2002). In contrast, Stock and Watson (2003) conclude that stock returns have only marginal content for predicting output growth, although the results of Estrella and Mishkin (1998)... |

193 | Predicting U.S. Recessions: Financial Variables as Leading Indicators - Estrella, Mishkin - 1998 |

174 | State-space models with regime switching - Kim, Nelson - 1999 |

139 | Testing linearity against smooth transition autoregressive models - Luukkonen, Saikkonen, et al. - 1988 |

111 | What does the yield curve tell us about GDP growth
- Ang, Piazzesi, et al.
- 2006
(Show Context)
Citation Context ... when K = m: yt+h = 0;h + 1;hxt + "t+h: (2) This type of regression was employed to measure the ability of the spread to predict output growth (Estrella and Hardouvelis, 1991; Hamilton and Kim, 2002; =-=Ang et al., 2006-=-), of dividend/price ratios for excess returns (see Cochrane (2005), ch. 20 for a survey), and of economic fundamentals for exchange rates (Kilian and Taylor, 2003). 2.2 Smooth Transition MIDAS Switch... |

102 | 2002a, “Regime Switches in Interest Rates
- Ang, Bekaert
(Show Context)
Citation Context ...s of non-linearities in …nancial time series, I use a threshold autoregressive model for generating x (m) t . Models with switching regimes have been employed previously for modelling interest rates (=-=Ang and Bekaert, 2002-=-). The two-regime threshold autoregressive model is estimated for the weekly data using p = 2m, that is, pX ! I(xt d c)+ (15) x (m) t = (2) 0 + (1) 0 + pX i=1 i=1 (1) i x(m) (1) t i + t (2) i x(m) (2)... |

97 | In-Sample or Out-of-Sample Tests of Predictability: Which One Should We Use?,” forthcoming, Econometric Reviews - Inoue, Kilian - 2004 |

90 | Tests for Forecast Encompassing - Harvey, Leybourne, et al. - 1998 |

88 | Were There Regime Switches in U.S. Monetary Policy - Sims, Zha - 2006 |

86 | 2006), “Predicting Volatility: How to Get the Most Out of Returns Data Sampled at Different Frequencies - Ghysels, Santa-Clara, et al. |

78 |
Testing and Modeling Threshold Autoregressive Processes
- Tsay
(Show Context)
Citation Context ...) x (m) t = (2) 0 + (1) 0 + pX i=1 i=1 (1) i x(m) (1) t i + t (2) i x(m) (2) t i + t ! I(xt d > c); where c is threshold and d is the delay. The parameters are estimated by conditional least squares (=-=Tsay, 1989-=-) using a grid search for the threshold and the delay. threshold has 100 points with the limits given by cL and cU The grid for the de…ned such that there are at least 15% of the observations in each ... |

77 | A simple, positive semi-de¯nite, heteroskedasticity and autocorrelation consistent covariance matrix - Newey, West - 1987 |

77 |
Modeling economic relationships with smooth transition regressions
- Teräsvirta
- 1998
(Show Context)
Citation Context ...igh frequency variable to predict a variable only available at a lower sampling frequency. The new regression model combines a non-linear time series regression model - smooth transition regressions (=-=Teräsvirta, 1998-=-) - with a MIxed Data Sampling approach - MIDAS (Ghysels, Santa-Clara and Valkanov, 2004). The mixed sampling approach allows for the direct use of high frequency data, while the smooth transition all... |

75 | Exchange Rates and Monetary Fundamentals: What Do We Learn from LongHorizon Regressions - Kilian - 1999 |

61 | Econometric Foundations - Mittelhammer, Judge, et al. - 2000 |

60 |
Long-Horizon Regressions: Theoretical Results and Applications
- Valkanov
- 2003
(Show Context)
Citation Context ...gressions for measuring the ability of a predictor to forecast a variable at long horizons may be questioned because a persistent predictor may be spuriously correlated with the cumulated regressand (=-=Valkanov, 2003-=-). Another problem of in-sample analysis is that the tests for no predictive content of xt to yt+h may be oversized because of the properties of the Newey-West estimator (Ang and Bekaert, 2007). The r... |

55 | Smooth transition autoregressive models-a survey of recent developments - Dijk, Teräsvirta, et al. - 2002 |

50 | A reexamination of the predictability of economic activity using the yield spread
- Hamilton, Kim
- 2002
(Show Context)
Citation Context ...rvey, 1988; Stock and Watson, 2003). Indeed, one of the most popular leading indicators of the US growth is the spread between long-term and short-term interest rates (Estrella and Hardouvelis, 1991; =-=Hamilton and Kim, 2002-=-). In contrast, Stock and Watson (2003) conclude that stock returns have only marginal content for predicting output growth, although the results of Estrella and Mishkin (1998) suggest some power in p... |

42 | 2004), ”A comparison of direct and iterated multistep AR methods for forecasting macroeconomic time series,” Working paper available at http://www.wws.princeton.edu/m̃watson/wp.html Naes - Marcellino, Stock, et al. - 1993 |

39 | Testing for Linearity
- Hansen
- 1999
(Show Context)
Citation Context ...nt distribution of a t-statistic. The bootstrap procedure has also some similarities with the one employed for computing the empirical distribution of the Hansen sup-test for threshold non-linearity (=-=Hansen, 2000-=-). Under the null, the data generating process for yt is: yt = (zt zt 1) = + (zt 1 zt 2) + t; : (14) where zt is the log of real GDP in dollars. Data on zt are used for estimating the conditional mean... |

35 |
How Stable is the Predictive Power of the Yield Curve? Evidence from Germany and the United States
- Estrella, Rodrigues, et al.
- 2003
(Show Context)
Citation Context ...e regime of lower predictive ability has a higher frequency after 1980, so this is in agreement with the literature that …nds instability in the predictive ability of the spread with data up to 1998 (=-=Estrella et al., 2003-=-). However, periods of low predictive ability can be identi…ed as the high-growth recovery periods that occur after recessions. This association with business cycle regimes suggests that recurrent cha... |

30 |
2004(b), “On the Selection of Forecasting Models
- Inoue, Kilian
(Show Context)
Citation Context ... measurement of the ability of x (m) t in predicting yt+h can be evaluated in comparison to regressions (eq. 2) and autoregressive models. The comparison of in-sample …t may use information criteria (=-=Inoue and Kilian, 2006-=-), while the out-of-sample performance may be compared in terms of means of squared forecast error. 2 Ghysels, Santa-Clara and Valkanov (2006) use a beta function to compute the weights. For the short... |

28 | MIDAS regressions: Further results and new directions - Ghysels, Sinko, et al. - 2006 |

21 | 2005): Small Sample Properties of Forecasts from Autoregressive Models under Structural Breaks,Journal of Econometrics
- Pesaran, Timmermann
(Show Context)
Citation Context ...wever, when there is an17 autoregressive term, the sample is shorter with “rolling” scheme so it increases the parameter bias; thus, even with a break, “rolling” may not be the most adequate method (=-=Pesaran and Timmermann, 2005-=-). An advantage of “recursive” forecasting is the requirement of a large sample size in order to …nd strong evidence of changes in predictive ability such that it matters for out-of-sample forecasting... |

16 |
Forecasting output and ination: The role of asset prices
- Stock, Watson
- 2003
(Show Context)
Citation Context ...e basis of expectations about future dividends and interest rates. This forward-looking characteristic suggests that bond and stock returns should be useful predictors of output growth (Harvey, 1988; =-=Stock and Watson, 2003-=-). Indeed, one of the most popular leading indicators of the US growth is the spread between long-term and short-term interest rates (Estrella and Hardouvelis, 1991; Hamilton and Kim, 2002). In contra... |

15 | 2007a), Nonparametric Bootstrap Procedures for Predictive Inference Based on Recursive Estimation Schemes - Corradi, Swanson |

14 |
Evaluating Direct Multi-Step Forecasts. Econometric Reviews
- Clark, McCracken
- 2005
(Show Context)
Citation Context ... has 14 years. Therefore, any break towards no predictive content in the end-of-sample will imply less power for the out-of-sample evaluation in detecting any predictive content of a given regressor (=-=Clark and McCracken, 2005-=-b). 3 http://research.stlouisfed.org/fred2/.14 3.1 In-sample Results Following the literature (for example, Estrella and Hardouvelis (1991)), the dependent variable is yt+h = (400=h) [zt+h zt], where... |

14 | Linear models, smooth transition autoregressions, and neural networks for forecasting macroeconomic time series: A reexamination - Teräsvirta, Dijk, et al. - 2005 |

13 |
Stock Return Predictability: Is it There? Review of Financial Studies 20:651–707
- Ang, Bekaert
- 2007
(Show Context)
Citation Context ...ights. I use the STMIDAS regression to measure the ability of asset returns in predicting output growth. This approach builds on predictive regressions. On the one hand, some authors (Valkanov, 2003; =-=Ang and Bekaert, 2007-=-) have criticized long-run regressions to measure the predictive power when applied to highly persistent regressors that are correlated with autoregressive disturbances. On the other hand, Inoue and K... |

11 | Testing for asymmetry in the link between the yield spread and output
- Galbraith, Tkacz
- 2000
(Show Context)
Citation Context ...ssions is that they are designed for direct forecasting. Previous applications of non-linear time series models for verifying changes in the dynamic relationship between output growth and the spread (=-=Galbraith and Tkacz, 2000-=-; Anderson and Vahid, 2001; Galvão, 2006) have speci…ed models only for one-step-ahead forecasts. Iterated forecasts for longer horizons are then obtained by bootstrap. Another alternative for modelli... |

10 | Macroeconomic Forecasting with Mixed Frequency Data: Forecasting US output growth, mimeo, link: http://www2.warwick.ac.uk/fac/soc/economics/staff/faculty/clements/wp/mpcabg.pdf, download: 26 - Clements, Galvão - 2007 |

10 |
Why is it so di¢ cult to beat the random walk forecast of exchange rates
- Kilian, Taylor
- 2003
(Show Context)
Citation Context ... Hardouvelis, 1991; Hamilton and Kim, 2002; Ang et al., 2006), of dividend/price ratios for excess returns (see Cochrane (2005), ch. 20 for a survey), and of economic fundamentals for exchange rates (=-=Kilian and Taylor, 2003-=-). 2.2 Smooth Transition MIDAS Switching regimes are a popular way of modelling nonlinear dynamics in regressions by using piecewise linear regimes linked by a transition function (Tong, 1990). When t... |

9 |
Predicting the probability of a recession with nonlinear autoregressive leading indicator models
- Anderson, Vahid
- 2001
(Show Context)
Citation Context ...igned for direct forecasting. Previous applications of non-linear time series models for verifying changes in the dynamic relationship between output growth and the spread (Galbraith and Tkacz, 2000; =-=Anderson and Vahid, 2001-=-; Galvão, 2006) have speci…ed models only for one-step-ahead forecasts. Iterated forecasts for longer horizons are then obtained by bootstrap. Another alternative for modelling switching regimes is to... |

9 |
Misspecifed structural change, threshold, and markov-switching models
- Carrasco
(Show Context)
Citation Context ... available at higher frequencies than the one on output growth. Modelling recurrent changes over time is an alternative to modelling breaks. Although switching-regimes models may also capture breaks (=-=Carrasco, 2002-=-), there are economic reasons for adopting models with recurrent regimes. Changes in the predictive power of asset returns for output growth may be related to business cycle regimes. An inverted yield... |

9 |
The power of tests of predictive ability in the presence of structural breaks
- Clark, McCracken
- 2005
(Show Context)
Citation Context ... has 14 years. Therefore, any break towards no predictive content in the end-of-sample will imply less power for the out-of-sample evaluation in detecting any predictive content of a given regressor (=-=Clark and McCracken, 2005-=-b). 3 http://research.stlouisfed.org/fred2/.14 3.1 In-sample Results Following the literature (for example, Estrella and Hardouvelis (1991)), the dependent variable is yt+h = (400=h) [zt+h zt], where... |

8 |
Forecasts Of Economic Growth From The Bond And Stock Markets
- Harvey
- 1989
(Show Context)
Citation Context ... are set on the basis of expectations about future dividends and interest rates. This forward-looking characteristic suggests that bond and stock returns should be useful predictors of output growth (=-=Harvey, 1988-=-; Stock and Watson, 2003). Indeed, one of the most popular leading indicators of the US growth is the spread between long-term and short-term interest rates (Estrella and Hardouvelis, 1991; Hamilton a... |

6 | The MIDAS touch: Mixed Data Sampling regression - Ghysels, Santa-Clara, et al. - 2004 |

5 |
2006): “Structural break threshold VARs for predicting US recessions using the spread,”Journal of Applied Econometrics
- Galvao
(Show Context)
Citation Context ...ng. Previous applications of non-linear time series models for verifying changes in the dynamic relationship between output growth and the spread (Galbraith and Tkacz, 2000; Anderson and Vahid, 2001; =-=Galvão, 2006-=-) have speci…ed models only for one-step-ahead forecasts. Iterated forecasts for longer horizons are then obtained by bootstrap. Another alternative for modelling switching regimes is to make the regi... |

5 | There is a risk-return tradeo¤ after all - Ghysels, Santa-Clara, et al. - 2005 |

4 | Bounds for Inference with Nuisance Parameters Present Only under the Alternative - ALTISSIMO, CORRADI |

4 | 2002), Determining the number of regimes in threshold autoregressive model using smooth transition autoregression, 13th model selection and evaluation EC2 conference, Bologna. o + o + o
- Strikholm, Tersvirta
(Show Context)
Citation Context ...m) 2;h x(m) t(^;m) 2 + "t+h: (6)8 The null hypothesis is (m) 2;h = 0; assuming that ^ has been estimated under the null. This variable addition test has also power for detecting threshold linearity (=-=Strikholm and Teräsvirta, 2005-=-). A problem of applying this approach for forecasting regressions is that the properties of the test are derived assuming that "t+h is iid. It is only reasonable to assume that this is the case when ... |

3 | Forecast content and content horizons for some important macroeconomic time series - W, Tkacz - 2007 |

2 |
A sequential procedure for determining the number of regimes in a threshold autoregressive model
- Strikholm, Teräsvirta
- 2006
(Show Context)
Citation Context ... (m) 2;h x(m) t(^;m) 2 + "t+h: (6) The null hypothesis is (m) 2;h = 0; assuming that ^ has been estimated under the null. This variable addition test has also power for detecting threshold linearity (=-=Strikholm and Teräsvirta, 2006-=-). The testing procedure proposed by Becker and Osborn (2007) imposes that data of higher frequency are only important to identify regimes, so that they do not use estimates of the weighting function ... |

1 | Weighted smooth transition regressions - Becker, Osborn - 2007 |