RESEARCH UPDATE Spring 2021 - FEATURES PUBLICATIONS NEWS & EVENTS PEOPLE ANNOUCEMENTS - Banco de España
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
RESEARCH Spring 2021
UPDATE
FEATURES
PUBLICATIONS
NEWS & EVENTS
PEOPLE
ANNOUCEMENTSWelcome to the Banco
de España Research Update
The Banco de España is pleased to announce the release of the Spring 2021 issue of its
Research Update. The Update aims to inform both academic and policy-oriented
economists and financial specialists about publications, conferences, and other research
activities at the Banco de España, during the semester from October 2020 to April 2021.
As usual, this issue includes several feature articles summarizing policy-relevant findings
from recent Banco de España projects in diverse areas of research. First, J. Andrés, O.
Árce and P. Burriel analyse whow the dampening of the estimated co-movement between
the output gap and inflation noticed in recent decades can be rationalized through a model
that incorporates the increase in polarization observed in several industries. Second, D.
Serrano-Puente evaluates the optimal progressivity of the Spanish personal income tax
scheme through the lens of a general equilibrium model featuring rich household
heterogeneity. He finds that there is room for welfare improvement by means of modifying
the effective average tax rates, at the expense of efficiency losses. Third, A. Alonso and J.
M. Carbó study the economic impact for financial institutions of using machine learning
(ML) models for credit default predictions. They find that, although ML models outperform
traditional ones, more complex algorithms do not imply better predictions. Fourth, F. Holub,
L. Hospido and U. Wagner estimate the impact of air pollution on the incidence of sick
leaves on a representative panel of employees affiliated to the Spanish social security
system. Controlling for several possible confounding factors, their evidence points towards
an impact of air pollution on the incidence of sick leaves, especially for women with pre-
existing conditions. Finally, P. García-Perea, A. Lacuesta and P. Roldán-Blanco document
that the increase in mark-ups in Spain during the financial crisis can be explained by the
response of relatively small and unproductive firms to the drop in their sales and the increase
in average costs.
In addition, the Update reports on other research news such as conferences and recent
publications, including the Financial Stability Review, a half-year journal published by the
Banco de España. This issue also covers an interview with Enrique Moral-Benito, Head of
the Sectoral Analysis Division, covering his research agenda, the policy developments at
the Banco de España related to the ongoing Covid19 crisis, as well as recent efforts to
develop and leverage large and granular datasets in order to tackle research and policy
questions.
We highlight these and other research developments at the Banco de España in hopes that
they will interest the broader research community in Spain and internationally, and thereby
contribute to an improved understanding of economic policy
Óscar Arce
Olympia Bover
Ángel Estrada
Eva Ortega
Carlos Thomas
Research Committee,
Banco de EspañaFEATURES
Market polarization unprecedented set of non-conventional monetary policy
measures after the global financial crisis (GFC), inflation and
and the Phillips curve inflation expectations in most advanced economies
JAVIER ANDRÉS, ÓSCAR ARCE AND PABLO BURRIEL remained chronically subdued even before the Covid-19
Summary of Banco de España Working Paper no. 2106 crisis. Some argue that this comes from afar and potentially
well before the GFC (Blanchard, 2016). Prominent among
The Phillips curve has flattened out over the last decades. the potential explanations for the weak reaction of prices to
We develop a model that rationalizes this phenomenon as cyclical conditions are the decline of labor power, the rise of
a result of the observed increase in polarization in many globalization and international trade, and the impact of
industries, a process along which a few top firms gain an positive supply shocks caused by new technologies.
increasing share of their industry market. In the model,
firms compete à la Bertrand and there is exit and These factors do not exhaust the list of possible causes
endogenous market entry, as well as optimal up and behind the diminishing effect of cyclical fluctuations on
downgrading of technology. Firms with larger market prices. A related strand of literature is placing increasing
shares find optimal to dampen the response of their price attention on some ongoing significant changes in the
changes, thus cushioning the shocks to their marginal industrial structure in advanced economies. These changes
costs through endogenous countercyclical markups. include, among others, the rise in market shares in many
Thus, regardless of its causes (technology, competition, industries, industrial polarization along different dimensions
barriers to entry, etc.), the recent increase in polarization (e.g. firm size, productivity, etc.), the rise in markups,
in many industries emerges in the model as the key factor intensification of competition spurred by technology and the
in explaining the muted responses of inflation to decline of the labor share. In particular, the case for market
movements in the output gap witnessed recently. concentration over the last decades has been forcefully
established on empirical grounds (Covarrubias, Gutiérrez
and Philippon, 2019). Yet, so far little consensus can be
SUMMARY FOR THE APRIL RESEARCH UPDATE found about the likely effect of these factors on the inflation
rate or, more precisely, on the link between inflation and the
In this paper we set up a model consistent with many of economic slack (Van Reenen, 2018). The importance of the
the features highlighted by the empirical industrial new technological giants in shaping the way economies
organization literature, to argue that these changes might respond to shocks, the archetypal case being the
be behind the muted response of inflation to shocks to the disinflationary impact of Amazon, is in stark contrast with
marginal costs that has been observed lately. The flattening the long held view in mainstream macroeconomics,
of the Phillips curve that has taken place over the last according to which market concentration has been
twenty years poses a challenge for monetary authorities. considered a source of inflationary pressure.
On the other hand, the availability of large data sets
containing firm level information on a number of relevant Our model sheds light on the connection between the rise
variables has uncovered a series of facts that speak of in market polarization (increase in market shares and
profound changes in the distribution of firms in many widening gap in size and productivity across firms in the
industries with repercussion in the response of same industry) and the flattening of the Phillips curve.
macroeconomic variables to exogenous shocks. Contrary to the previous standard view, in our model the
rise in market shares is neither inflationary nor deflationary
The pattern of inflation rates across most advanced per se, but it reduces the slope of the Phillips curve, which
economies in recent years defies the traditional explanations is consistent with the recent empirical literature (Del Negro,
based on the Phillips curve relation between inflation and Lenza, Primiceri, and Tambalotti, 2020). In particular, the
the output gap. In spite of the implementation of an slope of the Phillips curve derived in conventional New
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 3FEATURES
Keynesian DSGE models (regardless of whether price s
where ξN =
s
(
s ε −1 ) s
∂ξ
> 0, Ns > 0.
inertia is of the Calvo or the Rotemberg type) gets
1 − s ε 1 − s − 1
s s ∂s
augmented in our model by a factor that decreases with
the market share of the firm, as shown in equation (1).
Therefore, when a firm faces a positive shock to its
1 − s
θp 1 − βθ s marginal cost it raises prices, which in turn undermines its
p
+ mc s
s s
πt = βE t π t +1 s t market share and hence its desired markup; this dampens
θp
the inflationary effect of the shock. The strength of this
Standard NKPC effect increases with the firm’s market share. Hence, in an
economy featuring highly polarized industrial structures,
s 1 − θp
s s
+ s s − βθ s E s s (1) with a few large and many small competitors in each
t p t t +1
1 − s s ε 1 − s s − 1 θs
industry, much in line with the aforementioned recent
p
evidence, the response of inflation to shocks becomes
heterogeneity effect more muted than it would be in a similar economy with a
more balanced distribution of firms. In our simulations,
s s
where π t , represents inflation of firms of size s, mc t , their small variations in the drivers of market concentration
s
real marginal costs and st , their market share. 1
deliver significant changes in the slope of the Phillips
curve. While strategic price interactions barely affect the
Key to this result is the fact that markups depend positively markup of smallish firms, they do condition the desired
on the firm’s market share. The endogeneity of markups markup of large firms in a material manner (Amiti, Itskhoki,
hinges critically on the joint effect of two core features of the and Konings, 2019). This moderates the response of
industrial structure of the economy. We assume that firms prices set by the latter to shocks, which in turn (upon
have access to different TFP levels and choose among them aggregation) exerts a significant dampening effect on the
optimally taking into account the costs of moving up or down volatility of aggregate inflation. In fact, in the chart below
in the technology ladder. Furthermore, as in Etro and Rossi we show how a 20% increase in the productivity of larger
(2015), Andrés and Burriel (2018), and, more recently, Wang firms, which increases their market share, reduces the
and Werning (2020), we assume that firms compete à la response of their inflation to a negative TFP shock
Bertrand taking into account the expected reaction of other significantly (by 28%), while the response of smaller firms
competitors when setting their prices. Thus, substituting for to the same TFP shock remains unaltered. As a
the endogenous market shares in equation (1) and solving, consequence, the response of aggregate inflation is also
we get a Phillips Curve with the usual drivers of inflation, more moderate (by 26%). A similar result is obtained after
expected inflation and marginal costs, pre-multiplied by a an increase in the degree of competition as measured by
factor smaller than 1, which depends on the firms’ steady the elasticity of demand. Moreover, these findings are also
s
state market shares (s ), as shown in equation (2). found for other standard shocks, like an increase in policy
rates, or negative preference or labour supply shocks.
1 + θs ξ s 1 − θs 1 − βθ s
1 p p
s p N s mc s
πt = βE t π t +1 + To further assess the relevance of the mechanism analyzed
1 + ξNs s s t
1 + ξN θp
in the paper, we investigate the medium term response of
(2) the industrial structure to increases in technological
divergences, the elasticity of substitution among goods
and barriers to entry, three factors that Covarrubias,
1 In the model size and productivity level are interchangeable. The
s
Gutiérrez and Philippon (2019) identify as the main drivers
parameter 1 − θp represents the share of firms that are allowed to
change their prices every period according to the Calvo price-setting of market concentration in recent decades. The model
mechanism, β is the discount factor and ε the elasticity of substitution
between intermediate goods. predictions are consistent with the main facts reported by
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 4FEATURES
Chart 1
IMPULSE RESPONSE OF INFLATION AFTER AN INCREASE IN TFP UNDER DIFFERENT PRODUCTIVITY SCENARIOS
1 AGGREGATE INFLATION 2 INFLATION OF HIGH -TFP FIRMS 3 INFLATION OF LOW -TFP FIRMS
% % %
0.00 0.00 0.04
-0.01 -0.01 0.00
-0.02 -0.01 -0.04
-0.03 -0.02 -0.08
-0.04 -0.02 -0.12
-0.05 -0.03 -0.16
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Quarters Quarters Quarters
BASELINE 20% INCREASE IN HIGH-TFP FIRMS PRODUCTIVITY
a The variables are presented as percentage differences with respect to the steady state.
the empirical literature: (1) Sustained increase in Andrés, J. and P. Burriel, 2018): “Inflation and optimal monetary
policy in a model with firm heterogeneity and Bertrand
concentration in most industries (Bajgar, Berlingieri,
competition”. European Economic Review, 103, 2018, Pages
Calligaris, Criscuolo, and Timmis, 2019); (2) concentration 18-38
in employment but less intense than in sales (Autor, Dorn,
Autor, D., D. Dorn, L. Katz, C. Patterson, and J. Van Reenen
Katz, Patterson, and Van Reenen, 2019); (3) increase in (2019): “The Fall of the Labor Share and the Rise of Superstar
polarization along other dimensions like productivity Firms”, Quarterly Journal of Economics (forthcoming)
(Berlingueri, Blanchenay and Criscuolo (2017); (4) Bajgar, M., G. Berlingieri, S.Calligaris, C. Criscuolo, and J.
sustained increase of markups, mostly at the top of the Timmis (2019): “Industry Concentration in Europe and North
America,” CEP Discussion Papers dp1654, Centre for
markup distribution (De Loecker, Eeckhout, and Unger
Economic Performance, LSE.
(2020); (5) steady decline in firm entry in most industries
(Akcigit and Ates, 2019); (6) generalized fall in investment Barkai, S. 2019: “Declining labor and capital shares”, Journal of
Finance (forthcoming)
rates in many advanced economies (Eggertsson, Robbins,
and Wold (2018); and (7) decline in the labor share Blanchard, O. (2016): “The US Phillips Curve: Back to the 60s?”,
American Economic Review: Papers & Proceedings 106(5):
accompanied by a fall in capital share in most industries
31--34.
too (Barkai, 2019).
Covarrubias, M., G. Gutiérrez and T. Philippon (2019): “From
Good to Bad Concentration? U.S. Industries Over the Past 30
REFERENCES Years”, NBER, Working Paper 25983.
Akcigit, U, and S T Ates (2019), “Ten Facts on Declining Business De Loecker, J., J. Eeckhout, G. Unger, 2020. “The Rise of
Dynamism and Lessons from Endogenous Growth Theory,” Market Power and the Macroeconomic Implications”,
NBER Working Paper 25755. Quarterly Journal of Economics 135(2), 2020, 561-644.
Amiti, M., O. Itskhoki, and J. Konings, 2019: “International Del Negro, M., M. Lenza, G. E. Primiceri, and A. Tambalotti
Shocks, Variable Markups, and Domestic Prices,” The Review (2020): “What’s up with the Phillips Curve?,” NBER Working
of Economic Studies, 02. Papers 27003
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 5FEATURES Eggertsson, G., J. Robbins, and E. Wold, 2018: “Kaldor and Piketty’s Facts: The Rise of Monopoly Power in the United States”, NBER Working Paper No. 24287 Etro, F. and L. Rossi, 2015. “New-Keynesian Phillips curve with Bertrand competition and endogenous entry”, Journal of Economic Dynamics & Control, vol. 51(3), pp. 318--340. Van Reenen, J., 2018. “Increasing Differences Between Firms: Market Power and the Macro-Economy,” CEP Discussion Papers DP1576, Centre for Economic Performance, LSE. Wang, O. and I. Werning, 2020. “Dynamic Oligopoly and Price Stickiness,” NBER Working Papers 27536, National Bureau of Economic Research, Inc. BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 6
FEATURES
Optimal Progressivity of Personal economies. For that reason, the optimal design of a
redistributive tax system is subject to many constraints, as
Income Tax: A General Equilibrium argued by Bakis et al. (2015). This is why having a
Evaluation for Spain quantitative theory that accounts accurately for the observed
income and wealth inequality is crucial when assessing the
DARÍO SERRANO-PUENTE
aggregate, distributional, and welfare implications of certain
Summary of Banco de España Working Paper no. 2101
policies. For that purpose, a heterogeneous households
Published in SERIEs - Journal of the Spanish Economic
Association, vol. 11(4), pp. 407-455, November 2020 general equilibrium model is here calibrated to replicate
some characteristics of the Spanish economy and used to
Is the Spanish economy positioned at its optimal compare the steady-state consequences of setting an
progressivity level in personal income tax? This article optimal progressivity level in the Spanish personal income
quantifies the aggregate, distributional, and welfare tax. This frames the setup here presented in literature of
consequences of moving toward such an optimal level. A general equilibrium models with heterogeneous agents to
heterogeneous households general equilibrium model explore the relationship between fiscal policy variables and
featuring both life cycle and dynastic elements is calibrated the endogenous cross-sectional distribution of income
to replicate some characteristics of the Spanish economy and wealth in Spain. Examples of this body of literature are
and used to evaluate potential reforms of the tax system. Pijoan-Mas and González-Torrabadella (2006), Díaz-
The findings suggest that increasing progressivity would Giménez and Pijoan-Mas (2019), and Guner et al. (2020),
be optimal, even though it would involve an efficiency loss. among others.
The optimal reform of the tax schedule would reduce
wealth and income inequality at the cost of negative effects The theoretical framework of this paper is built for Spain
on capital, labor, and output. Finally, these theoretical following Castañeda et al. (2003). The model is devoted to
results are evaluated using tax microdata and describe a (i) account for income and wealth inequality and (ii) study
current scenario where the income-top households decisions of households that face labor income processes
typically face suboptimal effective average tax rates. that are random, household-specific, and uninsurable. In
these model-based economies, households accumulate
wealth in part to smooth their consumption. Heterogeneity
Many modern governments implement a redistributive fiscal is introduced in this setup via distinct labor market
policy, where personal income is taxed at an increasingly opportunities using an uninsurable process on the
higher rate, while transfers tend to target the poorest endowment of efficiency labor units that features non-linear
households. The taxation of personal income is not a minor dynamics. Given the labor market opportunity, the
issue, since most of the OECD economies obtain a large households choose their work effort. In other words,
proportion of their tax collection through it. Raising taxes on
1
the labor choice is set here to be endogenous. Life cycle
higher incomes may be potentially justified by the increase in characteristics are modeled using aging and retirement and
income and wealth inequality in recent years in Spain, dynastic links are modeled in a way that households are
especially after the 2007 crisis, as documented by Angel et altruistic toward their descendants (hybrid model with
al. (2018). Beforehand, one is likely to consider that raising retirement and bequests). Household face a progressive tax
taxes on the income-rich households could reverse the schedule modeled through the Heathcote et al. (2017)
growing concentration at the top. However, this type of policy specification, a function that allows for assessing average
could be very costly in terms of efficiency in advanced market level of taxes and progressivity separately. The model is
properly calibrated to match some empirical statistics of the
Spanish economy and replicates the distributions of income
The OECD average of the share of personal income tax
1 and wealth in very much detail (also at the very top tails of
revenues over total tax revenues (excluding social security
contributions) has been around 30-35% in recent years. those distributions).
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 7FEATURES
Once the theoretical framework is defined, a bunch of potential decomposed by household type, where it is observed that the
progressivity reforms are assessed through the calculation of poorest working and non-working households are the ones
many different general equilibria (one equilibrium for each who benefit the most from the reform. Contrarily, the most
degree of progressivity evaluated). Then a Benthamite social efficient working households and the wealthiest ones (either
planner, who takes into account all households in the economy working or non-working) are those who experience the largest
by putting the same weight on each of them, discerns the trade-off between (i) positive welfare effects derived from higher
optimal progressivity reform that leaves the aggregate tax income (due to an increased interest rate that pushes up
burden and transfers-to-output levels unchanged. The findings capital returns) and (ii) adverse effects emerging from higher
suggest that aggregate social welfare is maximized when the tax payments (due to the increase in progressivity of the
level of progressivity of the Spanish personal income tax is income tax that discourages labor and savings). The losses
increased to some extent. More precisely, in the optimally from this trade-off are particularly high in top parts of the
reformed scenario (setting the optimal level of progressivity), income and wealth distributions and clearly offset the potential
welfare gains are equivalent to an average increase of 3.08% welfare gains of the households populating such areas.
of consumption. Therefore, knowing that these agents would be the losers
of the reform, despite positive aggregate welfare effects,
By decomposing the aggregate welfare change, it is shown the consequences on aggregate capital, labor, and
that most of the welfare gains are obtained from a majority of output would be negative, which means that the economy
households facing a lower tax rate, i.e. the poorest households would experience an efficiency loss. Moreover, looking at
facing lower effective income tax rates and richest households the distributional implications, this reform would reduce
affronting higher effective income tax rates. On the contrary, income and wealth inequality.
the general equilibrium effects of the optimal reformed
economy (higher interest rate and lower wage) and the effects Finally, the theoretical results are evaluated with Spanish tax
resulting from changes in the equilibrium distribution of micro data. From the point of view of a Benthamite social
households across income levels (larger mass of households planner, households between the 20th and the 80th
at lower income levels) show a welfare loss, but these losses percentiles would experience a decrease in their average tax
are so small that together cannot overpass the welfare gains rates under the optimal progressivity reform. For example,
directly coming from the reformed tax system, jointly resulting the effective average tax rate encountered by a household
in positive aggregate welfare changes. These welfare gains are situated within the 40th and the 60th percentiles of the
Figure 1
AGGREGATE WELFARE CHANGE
% change in consumption
4
2
0
-2
-4
-6
-8
A ct u a l Optimal
-10
-12
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Progressivity level
AGGREGATE WELFARE CHANGE
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 8FEATURES
Figure 2
AVERAGE EFFECTIVE TAX RATES AND SHARE OF TAX REVENUES
1 AVERAGE EFFECTIVE TAX RATES
Average effective tax rates (%)
40
33.00
28.40
30
25.20
22.40
18.00 18.80 18.20 18.80
20
11.70 11.00
10 6.70
5.60
2.00 1.40
0.10 0.01
0
0-20 20-40 40-60 60-80 80-100 90-95 95-99 99-100
Percentiles of household gross income
2 SHARE OF TOTAL INCOME TAX REVENUE
Share of total income tax revenues (%)
100
90
78.13
80 74.88
70
60
50
40
26.47
30 20.89 22.48 23.73
17.25 15.7
20 13.78 13.71
6.48 5.29
10 1.34 0.88
0.06 0.0001
0
0-20 20-40 40-60 60-80 80-100 90-95 95-99 99-100
Percentiles of household gross income
ACTUAL OPTIMAL
income distribution would drop from 0.067 to 0.056, which addition, the reform would reduce income and wealth
involves a change of 1.1 p.p.. On the other hand, households inequality. However, this would lead to an efficiency loss
above the 80th percentile would experience a drastic increment of the economy, since it discourages work and savings
in their effective average tax rate. For instance, the top 1% mainly by penalizing the top-working and wealthiest
households of the gross income distribution would go from households.
confronting an average tax rate of 0.284 in the actual scenario
to dealing with an average tax rate of 0.330 in the optimal one. REFERENCES
Anghel, B., H. Basso, O. Bover, J. M. Casado, L. Hospido, M.
In conclusion, as policy implications arising from this Izquierdo, I. A. Kataryniuk, A. Lacuesta, J. M. Montero, and E.
study, what the model (jointly with the data) indicates is Vozmediano (2018): “Income, Consumption and Wealth
Inequality in Spain,” SERIEs - Journal of the Spanish Economic
that, in terms of aggregate welfare, it would be optimal to
Association - Journal of the Spanish Economic Association, 9,
increase the progressivity of the personal income tax. In 351–378.
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 9FEATURES Bakis, O., B. Kaymak, and M. Poschke (2015): “Transitional dynamics and the optimal progressivity of income redistribution,” Review of Economic Dynamics, 18, 679–693. Castaneda, A., J. Díaz-Giménez, and J. V. Ríos-Rull (2003): “Accounting for the U.S. Earnings and Wealth Inequality,” Journal of Political Economy, 111, 818–857. Díaz-Giménez, J. and J. Pijoan-Mas (2019): “Investment expensing and progressivity in flat-tax reforms,” SERIEs - Journal of the Spanish Economic Association, 10, 365–399. Guner, N., J. Lopez-Segovia, and R. Ramos (2020): “Reforming the Individual Income Tax in Spain,” SERIEs - Journal of the Spanish Economic Association, 11, 369-406. Heathcote J, Storesletten K, Violante G (2017) “Optimal tax progressivity: an analytical framework,” Quarterly Journal of Economics, 132, 1693–1754. Pijoan-Mas, J. and M. González-Torrabadella (2006): “Flat Tax Reforms: A General Equilibrium Evaluation for Spain,” Investigaciones Económicas, 30, 317–351. BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 10
FEATURES
Understanding the performance institutions could realize by using different ML algorithms.
While there exists an extensive and growing literature on
of machine learning models to the predictive gains of ML in credit default prediction,
predict credit default: a novel usually the findings are based on different sample sizes
and different types of underlying assets, making any
approach for supervisory conclusion not robust enough. Furthermore, the economic
evaluation impact of the use of ML in credit default prediction remains
understudied.
ANDRÉS ALONSO AND JOSÉ MANUEL CARBÓ
Summary of Banco de España Working Paper no. 2105
To tackle this research gap we use a unique and
We study the economic impact for financial institutions anonymized database provided by one of the most
of using machine learning (ML) models in credit default important Spanish banks. We first measure the relative
prediction. We do so by using a unique and anonymized performance of the following ML models, comparing it
database from a major Spanish bank. We first measure with a logistic regression (Logit): Lasso penalized logistic
the statistical performance in terms of predictive power, regression, Classification And Regression Tree (CART),
both in classification and calibration, comparing models Random Forest, XGBoost and Deep Neural Networks. To
like Logit and Lasso, with more advanced ones like this purpose we calculate the benefits in terms of statistical
Trees (CART), Random Forest, XGBoost and Deep performance assessing the predictive performance under
Learning. We find that ML models outperforms traditional different circumstances such as different sample sizes and
ones, although more complex ML algorithms do not different amount of explanatory variables. This allows us to
necessarily predict better. We then translate this into test whether the better statistical behavior of ML models
economic impact by estimating the savings in regulatory comes from an information advantage (associated to the
capital that an institution could achieve when using a access to big amounts of data) or model advantage
ML model instead of a simpler one to compute the risk- (associated to ML as high-end technology). We find that
weighted assets following the Internal Ratings Based ML models outperform Logit both in classification and in
(IRB) approach. Our benchmark results show that calibration, particularly XGBoost, existing a model
implementing XGBoost instead of Lasso could yield advantage that can be statistically isolated from an
savings from 12.4% to 17% in capital requirements, information advantage. Nevertheless, most complex
depending on the type of underlying assets. models like Deep Learning (Neural Networks), do not
necessarily predict better.
Recent surveys show that financial institutions are
increasingly adopting Machine Learning (ML) tools in Second, we propose a novel approach to translate this
several areas of credit risk management, like regulatory statistical performance into actual economic impact of
capital calculation, optimizing provisions, credit-scoring or using ML models in credit default prediction. Taking as a
monitoring outstanding loans (BoE, 2019; Fernández, basis the Basel formulas for risk-weighted assets (RWA)
2019). While ML models usually yield better predictive and the regulatory capital requirements in the Internal
performance, from a supervisory standpoint they also Ratings-Based (IRB) approach, we compute the savings in
bring new challenges, like interpretability of the results, terms of minimum capital requirements which could be
stability of the predictions and governance of the models achieved by using more advanced algorithms, in particular
(EBA, 2020; BdF, 2020). Given the novelty and complexity XGBoost, compared to traditional techniques like Lasso.
of some ML models, defining an adequate supervisory We perform a step-by-step computation of the capital
model evaluation approach is not an easy task. Therefore, requirements for both methods. Out of nearly 75,000 loans
before conducting any model risk analysis, it is essential to in our dataset, we use around 60,000 to train the models
understand the real economic gains that financial and make predictions of the probability of default (PD) over
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 11FEATURES
Figure 1
RANKING PDS PER MODEL
1 LASSO 2 XGBoss
0.30 0.35
0.25 0.30
0.25
0.20
0.20
0.15
0.15
0.10
0.10
0.05 0.05
0.00 0.00
1 8 15 22 29 36 43 50 1 8 15 22 29 36 43 50
OBSERVED DEFAULT RATE PD ESTIMATED
the remaining 15,000 loans.1 We organise the predictions 17 % lower for XGBoost than for Lasso. These capital
proportionally into 50 buckets (about 300 loans in each savings come from two sources. First, the difference in the
bucket), from lower to higher values of PD. The results are distribution of loans in buckets between models. Lasso’s
displayed in Figure 1. The discrepancy between the PD distribution is particularly flat in areas with low PD
observed default rate (blue line) and the average PD (red (Figure 1), accumulating a disproportionately large amount
line) is greater for Lasso than for XGBoost, as Lasso of loans at around 1.5% of PD. According to the Basel
tends to both overestimate and underestimate the formulas, the K function of a group of loans is mainly
fraction of default. concave and increases with the PD of the loans, particularly
for low PDs. Second, the difference in the number of
In order to get the approval from a supervisor, the buckets found within each model. Since XGBoost’s PD
classification into buckets must comply with two criteria: (i) distribution (Figure 1 right) fits the observed default better
risk heterogeneity between buckets, and (ii) risk than Lasso’s, XGBoost ends up with more buckets in the
homogeneity within buckets. To meet both criteria, we final rank (eight instead of six). This implies, due to the
sequentially reduce the number of buckets. Out of the 50 concavity of the RWA Basel function over the parameter
starting buckets, we end up with six for Lasso and eight PD, a difference in capital requirements in its favour.
for XGBoost. Lasso finds fewer buckets because we are
constrained by its underlying PD distribution, which Our results indicate that ML models, due to their better
presents important flat areas, undifferentiated, that do not statistical performance, could generate significant savings
allow further disaggregation (Figure 1 left). for financial institutions in terms of regulatory capital
requirements compared to traditional statistical models.
Once we have our final bucket classification for Lasso and The magnitude of our results suggests that supervisors
XGBoost, we calculate the capital requirements (K) for need to thoroughly investigate the risks associated with
each bucket, and find that the average K can be up to the use of these models, both from a micro and macro-
prudential perspective, in order to ease the adoption of
1 Different train-test partitions do not affect the results of this section. this innovation in the market.
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 12FEATURES
REFERENCES
BdF (2020). “Governance of Artificial Intelligence in Finance,”
Fintech Innovation Hub ACPR. Banque de France. June 2020.
BoE (2019). “Machine learning in UK financial services.” Bank of
England
Fernández, Ana (2019). ”Inteligencia artificial en los servicios
financieros.” Boletín Económico 2/2019. Artículos Analíticos.
Banco de España
EBA (2020). ”Report on Big Data and Advanced Analytics.”
European Banking Authority.
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 13FEATURES
Urban air pollution and sick observed in post-industrial economies. Our empirical analysis
is based on a novel administrative dataset from the period
leaves: Evidence from social 2005-2014 that links rich information on personal and
security data occupational characteristics of Spanish workers to the
frequency, length, and diagnosis, of sick leaves taken. We
FELIX HOLUB, LAURA HOSPIDO
AND ULRICH J. WAGNER estimate the impact of air pollution on workers’ propensity to
Summary of Banco de España Working Paper no. 2041
call in sick, based on weekly variation in ambient
concentrations of particulate matter (PM10) across 99 cities in
Air pollution poses a major threat to public health by Spain. Our baseline model is a linear regression of the share
shortening lives (Deryugina et al., 2019) and increasing acute of sick-leave days on the share of high-pollution days and
morbidity (Schlenker and Walker, 2016). Air pollution causes weather variables. To control for non-random assignment of
additional damage by reducing productivity on the job (Graff pollution across workers, we include city-by-year, year-by-
Zivin and Neidell, 2012) and by hindering human capital quarter, and worker fixed-effects. Possible remaining
accumulation (Currie et al., 2009; Ebenstein et al., 2016). endogeneity is addressed in an instrumental-variables (IV)
Recent research has provided credible evidence that air regression that exploits exogenous variation in PM10 driven by
pollution damages the economy also via reductions in labor dust storms in Northern Africa. Under certain meteorological
supply in the context of emerging economies (Hanna and conditions, storms in the Sahara Desert stir up dust into high
Oliva, 2015; Aragón et al., 2017), but little is known so far altitudes. These dust clouds can travel very long distances
about this relationship in post-industrial societies where and reach European territory several times a year. The arrival
pollution levels are low and productivity is high. of Sahara dust occurs throughout all of Spain, and it is most
frequently observed on the Canary Islands, due to their
Our paper provides the first causal estimates of how many geographical proximity to the Sahara, where the
work days are lost due to air pollution concentrations typically phenomenon is popularly known as “Calima”. Because
Figure 1
TRENDS IN POLLUTION, EMPLOYMENT AND WAGES 2005-2014
1 SHARE OF HIGH-POLLUTION DAYS (a) 2 EMPLOYMENT AND PRODUCER WAGES (b)
100
7.5
.4
.3
95
7
Million workers
Euros of 2018
.2
90
6.5
.1
85
0
6
2006 2008 2010 2012 2014 2006 2008 2010 2012 2014
Year Year
PM10 > 50 ?g/m³ Affiliated workers (cities in the sample)
PM10 > 37.5 ?g/m³ Real daily producer wage
NOTES: Figure 1.1 displays the share of worker days with PM10 concentration exceeding EU 24-hour limit of 50 mg/m3 (solid line) and the share of
worker days with concentration in excess of 75% of the limit value (dashed line). The figure is based on our sample of 99 Spanish cities with at least
40,000 inhabitants. PM10 concentrations are weighted by the number of social security affiliates in each city and year. Figure 1.2 displays the number
of workers affiliated with the General Social Security Regime on our sample of 99 Spanish cities with at least 40,000 inhabitants (solid line) and the
daily producer wage, expressed in constant 2018 Euros (dashed line).
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 14FEATURES
Calima events substantially increase non-anthropogenic response of vulnerable workers (defined as those belonging
PM10 concentrations, the cities affected by this to the top-five percentile of the distribution of sick leaves
phenomenon are allowed to discount the measured taken during the pre-estimation period) is more than three
24-hour-mean concentration for this effect. Official PM10 times stronger than the response of healthy workers.
discounts constitute a valid instrument for pollution Furthermore, our analysis reveals that job security matters, in
because they shift local PM10 concentrations in ways that that workers with a high predicted risk of losing their job
are plausibly orthogonal to local conditions that drive sick respond less strongly to a pollution shock than others. This
leaves, after conditioning on weather. finding suggests that exacerbated presenteeism could
adversely affect future health outcomes and lower productivity
Our IV estimates imply that a 10%-reduction in high-pollution in this tier of the labor market. These interactions of behavior
events reduces the weekly absence rate by 0.0213 and labor market institutions have a large impact on estimates
percentage points, i.e. by 0.8% of the mean absence rate of the external costs of air pollution that arise from changes in
(2.79%). The estimation results allow us to compute a lower labor supply. If we adjust the above-mentioned benefits of air
bound on the benefits of improving urban air quality in Spain. quality improvements to account for presenteeism, the
We proceed in two steps. First, we calculate the reduction in impact estimates imply a corresponding increase in
sick days caused by a specific improvement in air quality. To production worth €706 million. Irrespective of which number
translate this into a monetary benefit, we then multiply this one prefers, this exercise shows that the productivity-related
number by the average daily producer wage. Under the benefits of air quality improvements that occurred in Spain
assumption that workers are paid their marginal product, this between 2005 and 2014 were both economically and
approximates the value of incremental production enabled by statistically significant.
the reduction in sick days. While improving air quality yields
sizable additional benefits by reducing mortality, human REFERENCES
suffering, and medical treatment costs, we focus on foregone
Aragón, F., Miranda, J. J., and Oliva, P. (2017). Particulate matter
production because this component of the social costs of air and labor supply: The role of caregiving and non-linearities.
pollution is directly linked to our outcome variable and has not Journal of Environmental Economics and Management, 86,
295–309.
yet been quantified in previous research.
Currie, J., Hanushek, E. A., Kahn, E. M., Neidell, M., and Rivkin,
S. G. (2009). Does pollution increase school absences?
In particular, we evaluate the cumulative benefits of the
Review of Economics and Statistics, 91 (4), 682–694.
actual air quality improvements that have taken place in
urban Spain over the period. The left-hand side graph of Deryugina,T., Heutel, G., Miller, N. H., Molitor, D., and Reif, J.
(2019). The mortality and medical costs of air pollution:
Figure 1 shows that worker exposure to PM10 concentrations Evidence from changes in wind direction. American Economic
exceeding the EU 24-hour limit of 50 µg/m³ decreased from Review, 109 (12), 4178–4219.
18.8% in 2005 to just under 2% in 2014. This improvement Ebenstein, A., Lavy, V., and Roth, S. (2016). The long-run
in ambient air quality saved at least €503 million in foregone economic consequences of high-stakes examinations:
Evidence from transitory variation in pollution. American
production by reducing worker absence by more than 5.55
Economic Journal: Applied Economics, 8 (4), 36–65.
million days. It is important to note that this calculation is
affected by major economic fluctuations that occurred Graff Zivin, J. and Neidell, M. (2012). The impact of pollution on
worker productivity. American Economic Review, 102 (7),
during the sample period, depicted in the right-hand side 3652–3673.
graph of Figure 1. We account for this by using annual
Hanna, R. and Oliva, P. (2015). The effect of pollution on labor
values of employment and wages. supply: Evidence from a natural experiment in Mexico City.
Journal of Public Economics, 122, 68–79.
We uncover two important sources of treatment heterogeneity. Schlenker, W. and Walker, W. R. (2016). Airports, air pollution,
One relates to preexisting medical conditions that we infer and contemporaneous health. Review of Economic Studies,
from a worker’s sick leave record. We estimate that the health 83 (2), 768–809.
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 15FEATURES
Raising Markups to Survive: costs are a theoretical construct, so one needs a proxy for
them. Second, output price data at the firm level is scarce,
Small Spanish Firms during if not inexistent. For this reason, we rely on recent
the Great Recession methodological advances in production function
estimation (e.g. De Loecker and Warzynski, 2012),
PILAR GARCÍA-PEREA, AITOR LACUESTA
AND PAU ROLDAN-BLANCO measuring firm-level markups as the ratio of a variable
Summary of Banco de España Working Paper no. 2033
input’s elasticity to output on that input’s share of total
sales. The Banco de España data proves to be an excellent
Over the past few decades, profit rates and various other resource because not only is it highly representative of the
measures of market concentration have been on the rise in Spanish economy (containing a large amount of firms and
the United States (De Loecker et al, 2020). These patterns a very high percentage of aggregate value added), but it
of the data have attracted much attention recently in both also contains very disaggregated information on the
academic as well as policy circles for their impact on structure of costs, allowing us to disentangle variable
productivity dynamics, the composition of industries and inputs (e.g. materials and labor expenses from workers
ultimately consumer welfare. The evidence points to with fixed-term contracts) from fixed inputs (e.g. other
rapidly growing and highly productive firms as the drivers operating expenses and labor expenses from open-
of this phenomenon, with these firms leading the increase ended contracts with large firing costs). It is this latter
in markups and the decline in the labor share (Autor et al. aspect which allows us to make the main point of the
2017 and 2020). However, these trends seem to be less paper: the evolution of firm-level markups in Spain can be
pronounced in Europe, and be driven by factors seemingly explained by firms’ efforts to rebalance their cost structure
unrelated to the dynamics of superstar companies. between variable and fixed inputs, a behavioral response
to the cycle, rather than by reasons of a more structural
In this article, we exploit rich balance-sheet data from the nature. Therefore, in the case of Spain, the evolution of
Banco de España to document the behavior of markups markups may not reflect aggregate changes in the
at the aggregate level in Spain, and perform various competitive structure of markets, but rather an idiosyncratic
decompositions to help us identify the main drivers response of firms to economic conditions.
behind their evolution over the period 2004-2017. Our
main finding is that, contrary to the United States, markup The figure above shows the evolution of the sales-
dynamics were primarily led by small and unproductive weighted average markup (measured relative to materials)
firms. Particularly in response to the Great Recession of over the period 2004-2017. The average markup has been
2008, these firms were unable to increase their productive
efficiency when their average costs rose due to a sharp
Figure 1
increase in the fixed part of their production costs, and SALES-WEIGHTED AVERAGE MARKUP (2004 = 1)
this translated into higher price markups. This behavioral
1.14
response, related to the composition of the structure of
1.12
costs, seems to indicate that these firms, when faced
1.10
with economic hardship, may have preferred to increase
Normalized Markup
1.08
their markups in an attempt to survive in their sector even
1.06
if this may have come at the expense of losing some
1.04
market share in the process.
1.02
1.00
Measuring price markups (defined as the ratio of a firm’s
0.98
2004 2006 2008 2010 2012 2014 2016
final output price to the marginal cost of producing it) is
Year
challenging because of at least two reasons. First, marginal
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 16FEATURES
normalized to one in 2004. We observe that markups are in some specific sectors after 2009, most notably in
countercyclical, rising by about 13ppt between 2004 and Construction, Supplies and Real Estate.
2009, and declining slightly thereafter. The behavior in
the first part of the sample is roughly common across Most importantly, we find that the behavior of average
most sectors of activity, albeit markups continued to rise markups is driven by small and unproductive firms. First,
Figure 2
SALES-WEIGHTED AND UNWEIGHTED AVERAGE INPUT COSTS SHARES, RELATIVE TO FIRM SALES, BY TYPE OF INPUT
1 MATERIALS/SALES 2 GENERAL COSTS/SALES
0.65 0.28
0.26
0.60
0.24
0.55 0.22
0.20
0.50
0.18
0.16
0.45
0.14
0.40 0.12
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
3 LABOUR COSTS/SALES 4 TEMP LABOUR COSTS/SALES
0.40 0.09
0.08
0.35
0.07
0.30
0.06
0.25 0.05
0.04
0.20
0.03
0.15 0.02
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
5 PERM LABOUR COSTS/SALES 6 FINANCIAL EXPENSES/SALES
0.35 0.030
0.30 0.025
0.25 0,020
0.20 0.015
0.15 0.010
0.10 0.005
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
SALES-WEIGHTED UNWEIGHTED
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 17FEATURES
most of the increase during the Great Recession is due to However, we find that most of the evolution of the sales-
the firms at the very top of the markup distribution, who weighted markup is explained by changes in the composition
happen to be firms with relatively low levels of productivity of industries among surviving firms, as well as by changes in
and low market shares. Second, the firms that increase their the markups of these firms themselves, rather than by the
markup the most (in relative terms) during this period were extensive margin of entry and exit of firms.
precisely this group of small and unproductive companies.
All in all, our study demonstrates that understanding the
To understand what is behind this phenomenon, we next structure of firm’s balance sheets is key to understand
turn to the behavior of different variable and fixed costs over the behavior of markups in Spain. Rather than reflecting
this period. In the figure, the black line is the sales-weighted change in market power per se, the rise in markups during
share of each input relative to firm turnover (sales), while the the Great Recession seems to have been driven by a
red line is an unweighted average. By construction, when reshuffling of cost expenditures away from variable and into
the sales-weighted average is above the unweighted fixed inputs. This was particularly the case among small and
average, smaller firms (in terms of sales) have a lower input unproductive firms, who may have behaved in this manner
share of sales. As seen in the figure, there exist sizable in order to soften the blow of the shock onto their profits,
differences in the terms of the cost structure of firms by firm and thereby increase their chances of survival.
size, especially regarding the contribution of materials and
labor expenses. First, larger firms devote a larger share of REFERENCES
their sales to paying for material inputs (the variable input
Autor, D., D. Dorn, L. F. Katz, C. Patterson and J. V. Reenen
relative to which our markup estimates are computed). In 2017). Concentrating on the fall of the labor share. American
response to the Great Recession, all firms decreased the Economic Review: Papers & Proceedings, 107 (5), 180 – 185.
share of their sales that pays for material expenses, but this — (2020). The fall of the labor share and the rise of superstar
ratio decreased disproportionally more for smaller firms, firms. Quarterly Journal of Economics, 135, 645 – 709.
explaining the stronger response in markups for these De Loecker, J., J. Eeckhout and G. Unger (2020). The rise of
firms. This phenomenon is reversed for inputs with higher market power and the macroeconomic implications. Quarterly
adjustment costs, such as general costs and labor Journal of Economics, 135 (2), 561 – 644.
expenses related to permanent workers on open-ended — and F. Warzynski (2012). Markups and firm-level export
contracts with high firing costs. In both cases, the share of status. American Economic Review, 102 (6), 2437 – 2471.
sales paying for these fixed inputs increased, indicating
that in response to the adverse economic shock,
continuing firms shifted their resources away from variable
costs into fixed ones, perhaps in an attempt to soften the
adverse effects of the shock on their economic profits and
their chances at survival.
To complement this analysis, we conclude on the evolution
of firm demographics and measures of market concentration
during this period. We show that, in spite of the
aforementioned behavioral responses at the firm level, we
do observe a decline in firm entry (from about 11% to about
7% only in the period 2007-2009) and a rise in firm exit (from
about 4% to 8% in the same period), as well as a mild
increase in the share of industry sales captured by the ten
largest firms (on average, from 47% in 2008 to 56% in 2017).
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 18PUBLICATIONS
Recent working papers WHO TRULY BEARS (BANK) TAXES?
EVIDENCE FROM ONLY SHIFTING
Link to Working Papers page STATUTORY INCIDENCE
GABRIEL JIMÉNEZ, DAVID MARTÍNEZ-MIERA AND
SPILLOVER EFFECTS IN INTERNATIONAL JOSÉ-LUIS PEYDRÓ
BUSINESS CYCLES Working paper no. 2040
MÁXIMO CAMACHO, MATÍAS PACCE AND GABRIEL
PÉREZ-QUIRÓS
Working Paper no. 2034 URBAN AIR POLLUTION AND SICK LEAVES:
EVIDENCE FROM SOCIAL SECURITY DATA
FELIX HOLUB, LAURA HOSPIDO AND ULRICH J
APPLICATION OF TEXT MINING TO THE WAGNER
ANALYSIS OF CLIMATE-RELATED Working paper no. 2041
DISCLOSURES
ÁNGEL IVÁN MORENO AND TERESA CAMINERO
Working paper no. 2035
THE NARRATIVE ABOUT THE ECONOMY AS
A SHADOW FORECAST: AN ANALYSIS USING
BANCO DE ESPAÑA QUARTERLY REPORT
WAGE DETERMINATION AND THE BITE OF NÉLIDA DÍAZ SOBRINO, CORINNA GHIRELLI, SAMUEL
COLLECTIVE CONTRACTS IN ITALY AND HURTADO, JAVIER J. PÉREZ AND ALBERTO URTASUN
SPAIN: EVIDENCE FROM THE METAL Working paper no. 2042
WORKING INDUSTRY
EFFROSYNI ADAMOPOULOU AND ERNESTO
VILLANUEVA REFORMING THE INDIVIDUAL INCOME TAX
Working paper no. 2036
IN SPAIN
NEZIH GUNER, JAVIER LÓPEZ-SEGOVIA AND
ROBERTO RAMOS
SCREENING AND LOAN ORIGINATION TIME: Working paper no. 2043
LENDING STANDARDS, LOAN DEFAULTS
AND BANK FAILURES
MIKEL BEDAYO, GABRIEL JIMÉNEZ, JOSÉ-LUIS OPTIMAL PROGRESSIVITY OF PERSONAL
PEYDRÓ AND RAQUEL VEGAS INCOME TAX: A GENERAL EQUILIBRIUM
Working paper no. 2037
EVALUATION FOR SPAIN
DARÍO SERRANO-PUENTE
Working paper no. 2101
WHY COGNITIVE TEST SCORES OF SPANISH
ADULTS ARE SO LOW? THE ROLE OF
SCHOOLING AND SOCIOECONOMIC ECONOMIC UNCERTAINTY AND DIVISIVE
BACKGROUND POLITICS: EVIDENCE FROM THE
BRINDUSA ANGHEL, PILAR CUADRADO AND DOS ESPAÑAS
FEDERICO TAGLIATI SANDRA GARCÍA-URIBE, HANNES MUELLER
Working paper no. 2038 AND CARLOS SANZ
Working paper no. 2102
THE SHORT- AND LONG-RUN EMPLOYMENT
IMPACT OF COVID-19 THROUGH THE EMU DEEPENING AND SOVEREIGN DEBT
EFFECTS OF REAL AND FINANCIAL SPREADS: USING POLITICAL SPACE TO
SHOCKS ON NEW FIRMS ACHIEVE POLICY SPACE
CHRISTOPH ALBERT, ANDREA CAGGESE AND IVÁN KATARYNIUK, VÍCTOR MORA-BAJÉN AND
BEATRIZ GONZÁLEZ JAVIER J. PÉREZ
Working paper no. 2039 Working paper no. 2103
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 19PUBLICATIONS
ARE WE MOVING TOWARDS AN ENERGY- TIME VARIATION IN LIFECYCLE CONSUMPTION
EFFICIENT LOW-CARBON ECONOMY? AND INCOME
AN INPUT-OUTPUT LMDI DECOMPOSITION YUNUS AKSOY, HENRIQUE S. BASSO AND CAROLYN
OF CO2 EMISSIONS FOR SPAIN AND THE EU28 ST AUBYN
DARÍO SERRANO-PUENTE Working paper no. 2111
Working paper no. 2104
PRICE SETTING IN CHILE: MICRO EVIDENCE
UNDERSTANDING THE PERFORMANCE OF FROM CONSUMER ON-LINE PRICES DURING
MACHINE LEARNING MODELS TO PREDICT THE SOCIAL OUTBREAK AND COVID-19
CREDIT DEFAULT: A NOVEL APPROACH FOR JENNIFER PEÑA AND ELVIRA PRADES
SUPERVISORY EVALUATION Working paper no. 2112
ANDRÉS ALONSO AND JOSÉ MANUEL CARBÓ
Working paper no. 2105
DOES THE ADDED WORKER EFFECT MATTER?
NEZIH GUNER, YULIYA A. KULIKOVA AND ARNAU
MARKET POLARIZATION AND THE PHILLIPS VALLADARES-ESTEBAN
CURVE Working paper no. 2113
JAVIER ANDRÉS, ÓSCAR ARCE AND PABLO
BURRIEL
Working paper no. 2106
UNEQUAL TRADE, UNEQUAL GAINS: THE
HETEROGENEOUS IMPACT OF MERCOSUR
RODOLFO G. CAMPOS AND JACOPO TIMINI
NEW DIMENSIONS OF REGULATORY Working paper no. 2114
COMPLEXITY AND THEIR ECONOMIC COST.
AN ANALYSIS USING TEXT MINING
JUAN DE LUCIO AND JUAN S. MORA-SANGUINETTI
Working paper no. 2107
Recent occasional papers
Link to Occasional Papers page
ENDOGENOUS TIME VARIATION IN VECTOR TENDENCIAS RECIENTES DE LA POBLACIÓN EN
AUTOREGRESSIONS LAS ÁERAS RURALES Y URBANAS DE ESPAÑA
DANILO LEIVA-LEON AND LUIS UZEDA EDUARDO GUTIÉRREZ, ENRIQUE MORAL-BENITO
Working paper no. 2108 AND ROBERTO RAMOS
Occasional Paper no. 2027
WEATHER, MOBILITY AND THE EVOLUTION
OF THE COVID-19 PANDEMIC THE EFFECTS OF CHANGES IN THE
CORINNA GHIRELLI, ANDREA GONZÁLEZ, JOSÉ LUIS COMPOSITION OF EMPLOYMENT ON EURO
HERRERA AND SAMUEL HURTADO AREA WAGE GROWTH: PANEL DATA ANALYSIS
Working paper no. 2109 ÁNGEL LUIS GÓMEZ
Occasional Paper no. 2028
FED COMMUNICATION ON FINANCIAL
STABILITY CONCERNS AND MONETARY ANALYSIS OF INSOLVENCY PROCEEDINGS
POLICY DECISIONS: REVELATIONS IN SPAIN AGAINST THE BACKDROP OF THE
FROM SPEECHES COVID-19 CRISIS: INSOLVENCY PROCEEDINGS,
KLODIANA ISTREFI, FLORENS ODENDAHL PRE-INSOLVENCY ARRANGEMENTS AND
AND GIULIA SESTIERI THE INSOLVENCY MORATORIUM
Working paper no. 2110 MIGUEL GARCÍA-POSADA GÓMEZ
Occasional Paper no. 2029
BANCO DE ESPAÑA | RESEARCH UPDATE, SPRING 2021 | 20You can also read
