Work in progress (... or somewhere in the pipeline)
- Tractable model of sustainable debt
- Common misconceptions about credit and its dynamics
- Some forecasting stuff (satellite models for bank stress-testing... including plausible projections for adverse scenarios)
- System priors for VAR models
- System priors for filtering/trend estimation through Tchebyshev time polynomials
Some recent work
Assessing House Prices in Canada: Borrowing Capacity and Investment Approach (with Michal Andrle)
IMF Working Paper /19/248, 2019
This paper is an extended analysis of Canadian house prices for 11 metropolitan areas using both the borrowing capacity approach but also the investment approach and discussion of macro-prudential policies (see also IMF WP/19/59).
[pdf]
Assessing House Prices with Prudential and Valuation Measures (with Michal Andrle)
IMF Working Paper /19/59, 2019
In this paper we provide tools for assessing the house prices and housing valuation. We use the concept of borrowing capacity and net-present value models to appraise the value of housing assets. The advantage of our approach over econometric procedures is clear motivation by economic and finance theory and lack of revision variability. The empirical application focuses on house prices in Prague, the Czech Republic.
[pdf]
Simultaneous Confidence Bands: Exploring Distribution of Impulse-Response Functions
draft, 2018 - PRELIMINARY AND INCOMPLETE
This note proposes a set of exploratory and inferential routines aiming at better understanding the behaviour of impulse-response functions that are generated by structural models. Sign-identified VAR models can be tackled by these methods as well. Some may want to read the note only to learn how to construct the narrowest simultaneous bands with user-specified coverage. However, the exploratory potential of the proposed methods goes far beyond this goal.
[pdf]
A Note on the Efficient Simulation in State Space Models
manuscript, 2016
This is a short methodological note on the paper by Chan and Jeliazkov (2009). The main point was to highlight a fading wisdom of mathematical equivalence between the Kalman smoother (mean square error approach) and the stacked least squares problem. The approach taken here to derive the mean and variance of the states might be easier to follow for some readers, in particular for those familiar with the ridge regression. Moreover, under the least squares paradigm many important extentions can be integrated into the estimation process simply by adding additional least-square objectives.
[pdf]
System Priors for Econometric Time Series (with Michal Andrle)
CNB Working Paper 1/2017 (older version: IMF Working Paper /16/231, 2016)
The paper introduces system priors to Bayesian econometrics. System priors are priors about high-level features of the economic or econometric model, be it IRFs, variance decompositions, frequency-response functions or others. This paper shows how to elicit interesting priors when estimating a time series model, in this case an AR(2). One of the priors on parameters is that the system is stable and at least 60% of its variance is coming from the business cycle frequencies. Such prior makes the prior IRF more economically-meaningful. Extensions to BVARs are relatively simple and are forthcoming.
[Article, pdf] [Presentation, pdf] [R code] [Journal version (published as: Econometrics with System Priors)]
A new measure of the Financial Cycle: Application to the Czech Republic (with Jakub Seidler and Petr Hlavac)
Eastern European Economics, 54 (4), 2016
We propose a simple measure of swings in the risk perceptions across time. Something your mother (and your boss) should have no difficulties to understand. The paper presents an easy-to-construct indicator of the financial cycle that can help distinguish between good and bad credit expansions. This is based on the simple idea that credit expansions associated with ever rising real estate prices, unsustainable debt-service capacity, loosening credit standards and overly optimistic capital markets are probably not innocuous and signal a build-up of systemic risks.
[Journal version | WP version | Short (non-tech) version] [R code]
The Impact of Financial Variables on Czech Macroeconomic Developments: An Empirical Investigation (with Tomáš Adam)
CNB Working paper 11, 2014
Using dynamic model averaging for VARs, this study aims at stressing (...ehm, at quite a length) two basic facts. First, the models with traditional macro variables perform fairly well during the tranquil periods, however they seem to provide inadaquate picture around the tipping points of the cycle and during recessions (i.e. in periods when a need for a fair assessment of the economy is the most urgent). The theory should thus provide more elaborate insights on episodic events and explain the role of the financial sector in them. Second, any measures based on the overall fit of the model should be avoided when deciding upon the optimal lag length or optimal values of hyperparameters, if the researcher is only interested in the forecasting performace of the subset of variables. In this case, marginal predictive densities of the subset provide some forecasting gains.
[pdf]
Some other work
(e.g. Journal publications, selected thematic articles that appeared in the Financial Stability Reports...) can be found at my RePec profile.
IMF Working Paper /19/248, 2019
This paper is an extended analysis of Canadian house prices for 11 metropolitan areas using both the borrowing capacity approach but also the investment approach and discussion of macro-prudential policies (see also IMF WP/19/59).
[pdf]
Assessing House Prices with Prudential and Valuation Measures (with Michal Andrle)
IMF Working Paper /19/59, 2019
In this paper we provide tools for assessing the house prices and housing valuation. We use the concept of borrowing capacity and net-present value models to appraise the value of housing assets. The advantage of our approach over econometric procedures is clear motivation by economic and finance theory and lack of revision variability. The empirical application focuses on house prices in Prague, the Czech Republic.
[pdf]
Simultaneous Confidence Bands: Exploring Distribution of Impulse-Response Functions
draft, 2018 - PRELIMINARY AND INCOMPLETE
This note proposes a set of exploratory and inferential routines aiming at better understanding the behaviour of impulse-response functions that are generated by structural models. Sign-identified VAR models can be tackled by these methods as well. Some may want to read the note only to learn how to construct the narrowest simultaneous bands with user-specified coverage. However, the exploratory potential of the proposed methods goes far beyond this goal.
[pdf]
A Note on the Efficient Simulation in State Space Models
manuscript, 2016
This is a short methodological note on the paper by Chan and Jeliazkov (2009). The main point was to highlight a fading wisdom of mathematical equivalence between the Kalman smoother (mean square error approach) and the stacked least squares problem. The approach taken here to derive the mean and variance of the states might be easier to follow for some readers, in particular for those familiar with the ridge regression. Moreover, under the least squares paradigm many important extentions can be integrated into the estimation process simply by adding additional least-square objectives.
[pdf]
System Priors for Econometric Time Series (with Michal Andrle)
CNB Working Paper 1/2017 (older version: IMF Working Paper /16/231, 2016)
The paper introduces system priors to Bayesian econometrics. System priors are priors about high-level features of the economic or econometric model, be it IRFs, variance decompositions, frequency-response functions or others. This paper shows how to elicit interesting priors when estimating a time series model, in this case an AR(2). One of the priors on parameters is that the system is stable and at least 60% of its variance is coming from the business cycle frequencies. Such prior makes the prior IRF more economically-meaningful. Extensions to BVARs are relatively simple and are forthcoming.
[Article, pdf] [Presentation, pdf] [R code] [Journal version (published as: Econometrics with System Priors)]
A new measure of the Financial Cycle: Application to the Czech Republic (with Jakub Seidler and Petr Hlavac)
Eastern European Economics, 54 (4), 2016
We propose a simple measure of swings in the risk perceptions across time. Something your mother (and your boss) should have no difficulties to understand. The paper presents an easy-to-construct indicator of the financial cycle that can help distinguish between good and bad credit expansions. This is based on the simple idea that credit expansions associated with ever rising real estate prices, unsustainable debt-service capacity, loosening credit standards and overly optimistic capital markets are probably not innocuous and signal a build-up of systemic risks.
[Journal version | WP version | Short (non-tech) version] [R code]
The Impact of Financial Variables on Czech Macroeconomic Developments: An Empirical Investigation (with Tomáš Adam)
CNB Working paper 11, 2014
Using dynamic model averaging for VARs, this study aims at stressing (...ehm, at quite a length) two basic facts. First, the models with traditional macro variables perform fairly well during the tranquil periods, however they seem to provide inadaquate picture around the tipping points of the cycle and during recessions (i.e. in periods when a need for a fair assessment of the economy is the most urgent). The theory should thus provide more elaborate insights on episodic events and explain the role of the financial sector in them. Second, any measures based on the overall fit of the model should be avoided when deciding upon the optimal lag length or optimal values of hyperparameters, if the researcher is only interested in the forecasting performace of the subset of variables. In this case, marginal predictive densities of the subset provide some forecasting gains.
[pdf]
Some other work
(e.g. Journal publications, selected thematic articles that appeared in the Financial Stability Reports...) can be found at my RePec profile.