Online learning of time-varying stochastic factor structure by variational sequential Bayesian factor analysis

Investment tasks include forecasting volatilities and correlations of assets and portfolios. One of the tools widely utilized is stochastic factor analysis on a set of correlated time-series (e.g. asset returns). Published time-series factor models require either sufficiently wide time windows of observed data or numeric solutions by simulations. We developed a ‘variational sequential Bayesian factor analysis’ (VSBFA) algorithm to make online learning of time-varying stochastic factor structure. The VSBFA is an analytic filter to estimate unknown factor scores, factor loadings and residual variances. The covariance matrix of the time-series predicted by the VSBFA can be decomposed into loadings-based covariance and specific variances, and the former can be expressed by ‘explanatory factors’ such as systematic components of various financial market indices. We compared the VSBFA with the most practiced factor model relying on wide data windows, the rolling PCA (principal components analysis), by applying them to 9-year daily returns of 200 simulated stocks with the ‘true’ daily data-generating model completely known, and by using them to forecast volatilities of long-only and long/short global stock portfolios with 25-year monthly returns of more than 800 stocks worldwide. Accuracy of the forecast covariance matrices is measured by a (symmetrized) Kullback–Leibler distance, and accuracy of the forecast portfolio volatilities is measured by bias statistic, log-likelihood, Q-statistic, and portfolio volatility minimization. The factor-based covariance and specific variances predicted by the best VSBFA are significantly more accurate than those by the best rolling PCA.     

The dynamic nature of survival determinants in e-commerce

The dynamic effects of survival determinants in e-commerce are tested using longitudinal data on 460 e-tailers. This is achieved through the incorporation of both time-varying covariates and coefficients in a discrete hazard rate model. The model includes elements of competitive strategy, industry structure, firm and product characteristics, and the macro environment. The study demonstrates the changing effect over time of factors affecting survival. For example, order of entry advantages are observed, but they are short-lived. This finding shows that e-tailers cannot rely on early entry as a strategic move in the long run. E-tailers with more media presence seem to survive longer. Being publicly traded and selling products with digital characteristics present advantages for e-tailers only in the beginning years, but they are not sustainable over long time periods. Survival chances decrease with higher competitive density, market growth rate, and equity market level at the time of entry. Conversely, economic growth tends to increase survival chances. The study also finds an inverted-U relationship between the hazard of exit and firm age. The conclusion section discusses the implications of the time-varying nature of survival determinants.     

Interdependence or contagion: A model switching approach with a focus on Latin America

Empirical research analysing contagion has become increasingly fragmented. Different definitions of contagion have resulted in different methods being deployed to analyse financial transmission channels. This paper devises a novel econometric strategy where the nature of interdependencies, magnitude of interdependencies and transmission channels selected for inclusion can change over time. We thus appeal to multiple definitions of contagion, distinguishing between: interdependence, contagion through interdependence and abrupt contagion through changing linkages. Using our approach we analyse different crisis episodes in Latin America. Results generally indicate interdependence not contagion during the currency crises of the 1990s and Argentine crisis of 1998–2002. During the global financial crisis, results indicate abrupt contagion from the US to Argentina and Brazil. Mexico, however, experiences contagion through existing interdependencies with the US. Results also show that macroeconomic and uncertainty channels play a role during different crises not just financial channels. By establishing whether or not different interdependencies and transmission channels are present during different crises our model switching approach provides new insights.     

A Production Model With Firm-Specific Temporal Variation in Technical Inefficiency: With Application to Spanish Dairy Farms

This paper introduces a stochastic frontier production model which accommodates firm-specific temporal variation in technical inefficiency. Unlike Cornwell, Schmidt and Sickles (1990), technical inefficiency is not modeled through the intercept of the production frontier, but through an error component model. The proposed model is a generalization of the Battese and Coelli (1992) model, which imposed a common temporal pattern upon all firms. In our application involving Spanish dairy farms, we find that the new model is preferred to the Battese and Coelli (1992) model on the basis of the likelihood ratio test. Results provide a new source of information on the different patterns of technical inefficiency change among the 82 farms in the sample.     

Time-varying impact of monetary policy shocks on US stock returns: The role of investor sentiment

This paper investigates how monetary policy shock affects the stock market of the United States (US) conditional on states of investor sentiment. In this regard, we use a recently developed estimator that uses high-frequency surprises as a proxy for the structural monetary policy shocks, which in turn is achieved by integrating the current short-term rate surprises, which are least affected by an information effect, into a vector autoregressive (VAR) model as an exogenous variable. When allowing for time-varying model parameters, we find that, compared to the low investor sentiment regime, the negative reaction of stock returns to contractionary monetary policy shocks is stronger in the state associated with relatively higher investor sentiment. Our results are robust to alternative sample period (which excludes the zero lower bound) and model specification and also have important implications for academicians, investors, and policymakers.     

Temporal causality between house prices and output in the US: A bootstrap rolling-window approach

This paper examines the causal relationships between the real house price index and real GDP per capita in the US, using the bootstrap Granger (temporal) non-causality test and a fixed-size rolling-window estimation approach. We use quarterly time-series data on the real house price index and real GDP per capita, covering the period 1963:Q1 to 2012:Q2. The full-sample bootstrap non-Granger causality test result suggests the existence of a unidirectional causality running from the real house price index to real GDP per capita. A wide variety of tests of parameter constancy used to examine the stability of the estimated vector autoregressive models indicate short- and long-run instability. This suggests that we cannot rely on the full-sample causality tests and, hence, this warrants a time-varying (bootstrap) rolling-window approach to examine the causal relationship between these two variables. Using a rolling window size of 28 quarters, we find that while causality from the real house price to real GDP per capita occurs frequently, significant, but less frequent, evidence of real GDP per capita causing the real house price also occurs. These results imply that while the real house price leads real GDP per capita, in general (both during expansions and recessions), significant feedbacks also exist from real GDP per capita to the real house price.     

Monetary policy and macroprudential policy: Rivals or teammates?

This paper sheds some light on situations in which monetary and macroprudential policies may interact (and potentially get into conflict) and contributes to the discussion about the coordination of those policies. Using data for the Czech Republic and five euro area countries we show that monetary tightening has a negative impact on the credit-to-GDP ratio and the non-risk-weighted bank capital ratio (i.e. a positive impact on bank leverage), while these effects have strengthened considerably since mid-2011. This supports the view that accommodative monetary policy contributes to a build-up of financial vulnerabilities, i.e. it boosts the credit cycle. On the other hand, the effect of the higher bank capital ratio is associated with some degree of uncertainty. For these and other reasons, coordination of the two policies is necessary to avoid an undesirable policy mix preventing effective achievement of the main objectives in the two policy areas.     

Industry co-movement and cross-listing: Do home country factors matter?

This paper investigates the co-movement of American depositary receipts (ADRs) and the industry returns of home and U.S. markets with a focus on industry co-movement by applying time-varying and constant copulas model specifications. We also examine the impacts of country-specific factors on the industry co-movement in cross-listing issue. Co-movements are found in relation to the industry returns and economic asymmetry. The industry co-movement of ADR is higher with home country than that with the U.S., supporting the fundamental view. The results pertaining to the ADR-home pairs are as follows: (1) during economic recessions, the factors of influencing industry co-movement include import and the number of listed companies; (2) in contrast, during economic booms, industry co-movement is significantly influenced by energy import, export, import, the number of listed companies, and high-tech export. The factors that impact the industry co-movement of ADR-US contain foreign direct investment, reserve, cash surplus, energy import, and the number of listed companies.     

Dynamic volatility modelling of Bitcoin using time-varying transition probability Markov-switching GARCH model

Bitcoin (BTC), as the dominant cryptocurrency, has attracted tremendous attention lately due to its excessive volatility. This paper proposes the time-varying transition probability Markov-switching GARCH (TV-MSGARCH) models incorporated with BTC daily trading volume and daily Google searches singly and jointly as exogenous variables to model the volatility dynamics of BTC return series. Extensive comparisons are carried out to evaluate the modelling performances of the proposed model with the benchmark models such as GARCH, GJRGARCH, threshold GARCH, constant transition probability MSGARCH and MSGJRGARCH. Results reveal that the TV-MSGARCH models with skewed and fat-tailed distribution predominate other models for the in-sample model fitting based on Akaike information criterion and other benchmark criteria. Furthermore, it is found that the TV-MSGARCH model with BTC daily trading volume and student-t error distribution offers the best out-of-sample forecast evaluated based on the mean square error loss function using Hansen’s model confidence set. Filardo’s weighted transition probabilities are also computed and the results show the existence of time-varying effect on transition probabilities. Lastly, different levels of long and short positions of value-at-risk and the expected shortfall forecasts based on MSGARCH, MSGJRGARCH and TV-MSGARCH models are also examined.     

Bond yields and debt supply: new evidence through the lens of a preferred-habitat model

This paper examines the responsiveness of bond yields to changes in debt supply. The preferred-habitat theory predicts a positive relation between the term spread and relative supply of longer term debt, and that this relation is stronger when risk aversion is high. To capture this effect, a time-varying coefficient model is introduced and applied to German bond data. The results support the theoretical predictions and indicate substantial time variation: under high risk aversion, yield spreads react about three times more strongly than when risk aversion is low. The accumulated response of term spreads to a one standard deviation change in debt supply ranges between 4 and 46 basis points.