Central European Journal of Economic Modelling and Econometrics

The problem addressed in this article is predicting profitability of consumer loan portfolio. It is a topic of critical importance for the management of the lending business. The requirement to predict credit losses is also explicit in IFRS 9 and CECL accounting standards. This makes it also an interesting topic for financial auditors and banking supervision. This article proposes a method of estimating lifetime profitability of loans based on interpretable machine learning. The method can utilise a large set of input variables like socio-demographic, credit bureau or transactional data. It is compliant with IFRS 9 and CECL credit loss provisioning standards. Performance of proposed method is demonstrated on four portfolios extracted from real data of one of the consumer lending institutions operating on the Polish market. Satisfactory results were achieved in differentiating the portfolio from the perspective of expected loss and expected revenue. The quality of the profitability forecast was compared with a simple benchmark model, proving a superior quality of proposed model, especially when macroeconomic variables are added.

This study examines the quantile dependence and directional predictability from the US to a vast range of emerging equity markets (EMs). This issue is addressed using the cross-quantilogram approach and daily data from January 2004 to April 2025. Our findings confirm that extreme conditions in the US market affect various quantiles of stock returns in EMs; however, the detailed picture of the dependence patterns varies significantly between quantile orders. A large sharp decrease (increase) in US stock returns increases the likelihood of an immediate extreme low (high) return in EMs, and the directional predictability is much higher at the tails of the distribution than at the median. We also find evidence of an asymmetric effect across quantiles, with negative spillovers having a stronger impact than positive spillovers. Furthermore, a higher degree of connectedness is observed between the US and EMs in the Americas than in other regions. The predictability remains pronounced after controlling for different global uncertainty measures, such as the Volatility Index (VIX), Economic Policy Uncertainty (EPU), Equity Market-related Economic Uncertainty Index (EMU), and Geopolitical Risk Index (GPR). Notably, EMs exhibited stronger connections with the US during the 2008 Global Financial Crisis than during the COVID-19 pandemic or the Russia-Ukraine war.

The main aim of the paper is to compare the modelling ability of several alternative multiplicative error models for describing the dynamics of trade durations and intraday trading volumes. To formally compare the relative explanatory power of multiplicative error specifications, the Bayesian rules of comparing statistical models are applied. In the paper, we revisit Bayesian model comparison and compare the Bayes factors obtained using different approximations of the values of the marginal data densities. The Newton and Raftery's harmonic mean estimator, the corrected harmonic mean estimator of Pajor and Osiewalski, the corrected arithmetic mean estimator proposed by Pajor, and a standard Monte Carlo with importance sampling technique are approximations used. We consider twelve multiplicative error models that differ in the structure of the conditional mean equation and the distribution of innovations. The analysis considers models with the Burr and generalised gamma distributions for the error term. The MCMC methods are suitably adopted to obtain samples from the posterior densities of interest. The empirical part of the work includes modelling historical trade durations and more recent intraday trading volumes for selected equities on the Polish stock market.