Nowcasting GDP - A Scalable Approach Using DFM, Machine Learning and Novel Data, Applied to European Economies

Abstract

March 2022IMF Working Papers describe research in progress by the author(s) and are published to elicit comments and to encourage debate. The views expressed in IMF Working Papers are those of the author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

Nowcasting Madagascar's real GDP using machine learning algorithms

“…Within the context of Madagascar, the leading indicators listed in Table 1 are regarded as those capable of anticipating potential short-term fluctuations in the GDP. Moreover, these indicators are consistent with existing empirical research on real-time monitoring of economic activities [7,15,23,30,44,52,53].…”

“…Machine Learning algorithms have now become a valuable tool in economic modelling, demonstrating remarkable efficacy in the challenging task of nowcasting and forecasting GDP across diverse global contexts. This effectiveness is evident in advanced economies (e.g., Canada [54], China [67,69], Finland [26], Italy [21], Netherlands [42], New Zealand [55,56,60], South Africa [17], Sweden [40], USA [31,45], multiple European countries [23]), emerging markets and developing countries (e.g., Albania [66], Bangladesh [32], Belize and El Savador [5], Brazil [57], Egypt [1], Georgia [46], India [28], Indonesia [62], Lebanon [64], Malaysia [38], Peru [63])). Moreover, Machine learning algorithms are also proved to be very competitive with respect to standard econometric methods.…”

Nowcasting GDP: An Application to Portugal

“…First observations have already been made regarding using the Kalman filter in order to obtain a common factor that can be used as a predictor in the described bridge equations, eventually replacing the Economic Sentiment Indicator (ESI) or the Economic Climate Indicator (ICE). Another line of research could cover regularized regression methods (e.g., LASSO), regression trees, random forests, neural networks, reinforcement learning and other machine-learning techniques, following the seminal contributions of Mullainathan and Spiess [23] and Dauphin et al Dauphin et al [24]. Finally, MIDAS models [12] can be estimated using different parameterizations and lags.…”

Nowcasting inflation with Lasso‐regularized vector autoregressions and mixed frequency data

“…Even in this case, the forecast performances of our nowcasting strategy are tested considering an extended period spanning from July 2019 to August 2022. Reference to an extended sample allows us to verify whether model comparison results are sample-dependent, an issue that emerged in recent literature (de Bondt et al, 2021;Dauphin et al, 2022). To better highlight this point, the forecasting period over which model comparisons are evaluated is divided into three reference periods: a pre-COVID-19 period, spanning from July 2019 to February 2020; a COVID-19 crisis period, spanning from March 2020 to September 2021 (i.e., when the European mass vaccination campaign was almost completed, allowing for a generalized relaxation of non-pharmaceutical containment measures); a post-COVID-19/Energy crisis period, starting from October 2022 and still ongoing.…”