Nowcasting emerging market’s GDP: the importance of dimension reduction techniques

ABSTRACT

A number of recent studies in the macro-finance literature that addresses the link between asset prices and economic fluctuations have focused on the usefulness of various factor models in the context of now-casting using very big dataset. The issue of factor extraction is usually swept under the carpet in the factor model literature, where it seems that all that is needed is a large number of economic and financial variables. We contribute to this literature by analysing whether factor estimation methods matters for the performance of factor-based now-casting models based on selected emerging markets GDP. Ancillary findings based on our GDP now-casting experiments on major emerging market countries underscore the advantage of sparse principal component analysis-based factor estimation approach. These results show that imposing a sparse structure on the whole dataset is generally a useful step towards reducing the forecast errors in the context of GDP now-casting model specification.

KEYWORDS:

• Diffusion index
• factor extraction methods
• emerging markets
• factor model
• forecasting

JEL CLASSIFICATION:

• C53
• C61
• E37

Acknowledgments

The authors wish to thank Prof. Norman R. Swanson and Hyun Hak Kim for useful discussions and comments on earlier versions of this paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed here.

Notes

¹ A few of the relevant papers include those by Stock and Watson (2002) for USA, Schumacher (2010) for Germany, Luciani and Ricci (2014) for Norway and Cepni, Güney, and Swanson (2019) for emerging markets.

² For further discussion of factor estimation methods, see Schumacher (2007), Kim and Swanson (2014, 2018).

³ We choose the number of static and dynamic factors (r, q) and the lag order (p) in Eq.(2) by searching across all combinations of $r=1,...,4$, $q=1,...,4$ and $p=1,...,4.$. We report the selected (q, r, p) values by factor extraction method and country in Table A1 of online appendix.

⁴ See online appendix for details.

⁵ Augmented Lagrangian Method (ALM) has recently gained increasing attention due to their rapid convergence. See Naikal, Yang, and Sastry (2011) for details.

⁶ See Cardoso and Souloumiac (1993) for details.

⁷ Recall that there are seven forecast horizons and four countries, we have a total of 28 cases.