Labor Market Effects of Monetary Policy Across Workers and Firms

This paper uses Austrian social security records to analyze the effects of ECB monetary policy on the labor market. Our focus is on the role of worker and Ąrm wage-components, deĄned by an Abowd et al. (1999) wage regression. Our Ąndings show that monetary tightening causes the largest employment losses for low-paid workers who are employed in high-paying Ąrms before the tightening. Monetary tightening further causes a reallocation of workers to lower-paying Ąrms. In particular low-paid workers who were originally employed by low-paying Ąrms are prone to falling down the Ąrm wage ladder.


Introduction
The distributional effects of monetary policy are both of direct concern for policymakers and important for the transmission of monetary policy. 1 In fact, a growing empirical literature studies the distributional effects of monetary policy across workers and Ąrms.2However, understanding how the worker-level effects of monetary policy depend on both the worker type and the workerŠs Ąrm type remains largely unexplored.
A key aspect of worker and Ąrm heterogeneity is that they jointly determine the workerŠs wage.Wages depend on worker-speciĄc components (e.g., worker productivity) and Ąrm-speciĄc components (e.g., Ąrm proĄtability).Therefore, the distribution of workers across Ąrms matters for earnings inequality (e.g., Bagger and Lentz, 2018;Song et al., 2018;Bonhomme et al., 2019Bonhomme et al., , 2022)), productive efficiency (e.g., Hagedorn et al., 2017), and earnings losses (e.g., Gulyas and Pytka, 2019;Lachowska et al., 2020;Bertheau et al., 2022).In addition, worker and Ąrm type determine jointly whether a worker-Ąrm match is sustained.Importantly, it is ex ante unclear to what extent worker and Ąrm-speciĄc characteristics explain why some workers are more affected by monetary policy than others.In this paper, we empirically characterize the distributional effects of ECB monetary policy shocks across workers and Ąrms using Austrian social security records.Using an Abowd et al. (1999) wage regression, we estimate worker and Ąrm (wage) Ąxed effects.From a workerŠs perspective, the Ąrm Ąxed effect is arguably the most important aspect of Ąrm heterogeneity, as it measures the Ąrm wage premium relative to other Ąrms.We refer to workers with a high worker Ąxed effect as high-paid workers, and to Ąrms with a high Ąrm Ąxed effect as high-paying Ąrms, and analogously for low-paid workers and low-paying Ąrms.We document three novel results.First, we show that employment losses after monetary tightening are concentrated among low-paid workers in high-paying Ąrms.Second, monetary tightening increases the rate at which workers reallocate across Ąrms, in particular for lowpaid workers.Third, the Ąrms to which workers switch after monetary tightening tend to be lower-paying than their previous Ąrms.Especially low-paid workers who were originally employed in low-paying Ąrms reallocate to (even) lower-paying Ąrms.All results apply symmetrically to expansionary monetary policy.While our Ąnding that low-paid workers are more affected by monetary policy is in line with the previous literature (quoted above), the novelty of our results is the role of the workerŠs original employer for the distributional effects of monetary policy.As low-paid workers at high-paying Ąrms tend to become non-employed, low-paid workers at low-paying Ąrms tend to reallocate to lower-paying Ąrms.Although a large literature studies heterogeneous effects of monetary policy across workers or Ąrms, jointly studying worker and Ąrm heterogeneity has been largely ignored.An exception is Moser et al. (2022) which estimates the distributional effects of lower credit supply due to negative interest rates on employment and pay both within and between Ąrms.Another closely related paper is Crane et al. (2022) which studies the effects of recession across both worker and Ąrm ranks.Our analysis uses the universe of Austrian social security records, which includes a worker identiĄer, an establishment identiĄer, the start and end dates of employment and registered unemployment spells, the wage, and a few other worker characteristics.We use these records to construct a quarterly worker-level panel with 200 million observations between 1999 and 2018.We combine the worker panel with high-frequency identiĄed ECB monetary policy shocks (Altavilla et al., 2019;Jarociński and Karadi, 2020).To characterize the distributional effects of monetary policy, we estimate worker-level panel local projections.Our main Ąndings show statistically and economically signiĄcant heterogeneity in the employment effects of monetary policy across workers and Ąrms.Across all workers, the average employment probability is 0.27 percentage points (p.p.) lower one year after a one-standard deviation contractionary monetary policy shock, and the opposite for an expansionary shock.The average, however, masks large differences across workers.For workers with an abovemedian worker Ąxed effect, the employment probability falls by 0.23 p.p., while for workers with a below-median worker Ąxed effect the employment probability falls by 0.32 p.p.That is, low-paid workers are 40% more likely to become non-employed than high-paid workers.However, only examining the role of worker Ąxed effects misses large differences across Ąrm Ąxed effects.Perhaps surprisingly, among the low-paid workers, those originally employed at high-paying Ąrms are particularly likely to become non-employed.Their employment probability falls by 0.36 p.p. Conversely, the employment probability of low-paid workers at low-paying Ąrms only falls by 0.18 p.p. Monetary policy shocks not only affect the probability whether a worker is employed, but also induce reallocation of workers across Ąrms.On average, a one standard deviation monetary policy shock increases the likelihood of changing employers by 0.2 p.p. Job switching is especially concentrated among low-paid workers.These workers are three times more likely than high-paid workers to change employers in response to a monetary policy shock.A natural question that arises is where workers reallocate to: Are workers moving to better paying or worse paying employers?We Ąnd that across all workers switching employers, the average wage premium of Ąrms falls by 0.16% after a one-standard deviation contractionary monetary policy shock.In other words, workers reallocate to lower-paying Ąrms.Interestingly, this reallocation response is fairly similar when comparing low-paid to high-paid workers, and when comparing workers at low-paying to those at high-paying Ąrms.However, we do Ąnd large differences in the interaction of worker type and Ąrm type.In particular, we Ąnd that low-paid workers originally employed by low-paying Ąrms are disproportionately reallocating towards worse-paying Ąrms.In contrast, low-paid workers originally employed by high-paying Ąrms tend to reallocate to similar Ąrm types.Taken together, our results imply that contractionary monetary policy shocks especially hurt low paid workers across multiple dimensions.First, they lower their employment probability, especially for those originally employed at high-paying Ąrms.Second, even conditionally on re-employment, monetary policy induces a reallocation of low paid workers originally employed at worse-paying Ąrms to even worse-paying Ąrms.Our paper provides new empirical moments which can be useful for the further development of Heterogeneous Agent New Keynesian models.While our Ąndings highlight the role of both worker and Ąrm heterogeneity, existing models either feature only worker heterogeneity (e.g., Gornemann et al., 2021;Dolado et al., 2019;Bergman et al., 2022;Bhandari et al., 2021;Ravn and Sterk, 2020), or only Ąrm heterogeneity (e.g., Ottonello and Winberry, 2020;Meier and Reinelt, 2022).Instead, a New Keynesian model with two-sided heterogeneity would allow studying the positive and normative implications of our evidence.The paper is organized as follows: Section 2 describes the data.Section 3 provides evidence on the employment effects of monetary policy.Section 4 provides evidence on the reallocation effects of monetary policy.Section 5 provides a sensitivity analysis.Section 6 concludes.

Data
In this section, we describe the data and key variables used in our analysis.

Austrian Social Security Data
We use administrative data from the Austrian social security administration that cover the universe of administrative employment and unemployment records for all workers subject to social security from 1999 through 2018. 3The data include a worker identiĄer, an establishment identiĄer, the Ąrst and last day of employment and unemployment spells, the workerŠs age, and the establishmentŠs industry classiĄer.In the data, we observe only the establish-ment a worker is employed at, but not the Ąrm.At the same time, most establishments are owned by one-establishment Ąrms.For simplicity, we will refer to establishments as Ąrms in the remainder of the paper.For every worker-Ąrm match, we observe annual labor income.On average, we observe 2.7 million workers per year.We construct a worker panel based on which we estimate worker-level responses to monetary policy shocks.In theory, we could construct a daily panel, since both social security data and monetary policy shocks are available at a daily frequency.Such a panel, however, would include 20 billion observations rendering the regression analysis extremely burdensome if not infeasible.Furthermore, given the presence of various labor market frictions and the typically sluggish response of macroeconomic aggregates to monetary policy shocks we should not expect large employment responses at very short horizons.We therefore construct a quarterly worker panel.We focus on individuals with high labor force attachment by excluding workers below 26 and above 60 years old. 4ur sample only consists of employment spells subject to social security and registered unemployment spells.5There are several reasons why a worker may disappear from our sample.A worker may drop out of the labor force, move outside of Austria, or Ąnd employment not covered by social security such as self-employment.In our analysis, we have to take a stance on how to deĄne the employment status of workers who disappear from our dataset.We decide to only consider the employment and non-employment trajectories of workers who are either employed or registered as unemployed.We think of this choice as conservative, as we may underestimate the employment responses if workers are pushed outside of the labor force in response to monetary policy shocks. 6Our Ąnal panel has 213.9 million workerquarter observations and Table 1 provides summary statistics.As we use the universe of all employment observations subject to social security, the descriptive statistics mirror the labor market structure of Austria.

Worker and Firm Fixed Effects
Our goal in this paper is to empirically characterize the distributional effects of ECB monetary policy shocks across the joint distribution of worker and Ąrm types.We estimate worker and Ąrm types using the seminal Abowd et al. (1999) wage regression (in short: AKM).In particular, we estimate worker and Ąrm types through the Ąxed effects in the following annual wage regression where wage i,j,τ is the log daily wage of worker i, employed in Ąrm j in year τ , F j(i,τ ) is a Ąrm Ąxed effect, W i is a worker Ąxed effect, and X i,τ is a cubic polynomial of worker age.
For each worker and year, we select the dominant employer according to total yearly income.
Table 1 provides descriptive statistics of the worker and Ąrm Ąxed effects.The Ąrm Ąxed effect F j(i,τ ) for Ąrm j is assumed to be invariant over time and is identiĄed through wage changes of workers moving across Ąrms. 7Theoretically it is possible that the Ąrm Ąxed effect is affected by monetary policy shocks.Although monetary policy are at least an order of magnitude smaller in standard deviation than idiosyncratic shocks to Ąrms, to avoid endogeneity concerns, our analysis will mostly use the Ąrm and worker Ąxed effects estimated from a backward-looking 5-year rolling window.We denote the estimated worker and Ąrm Ąxed effects for the rolling windows by where the sample used to estimate W rolling i,τ and F rolling j(i,τ ),τ ranges from year τ − 4 to τ .To be able to compare the rolling-window estimates over time, we compute the percentile rank of these Ąxed effects, which we denote by When studying the reallocation of workers across Ąrms, we need a constant measure of Ąrm Ąxed effects over time.Thus, in Section 4 we will use the Ąrm Ąxed effects estimated in (2.1) over the entire sample.

ECB Monetary Policy Shocks
As ECB monetary policy shocks, we consider high-frequency changes in the Overnight Index Swap (OIS) rates around policy meetings of the ECB Governing Council.The OIS is a swap contract exchanging a Ąxed interest rate for the Ćoating Euro Overnight Index Average (Eonia) on the European interbank market.We exclusively consider scheduled meetings, which mitigates the problem that monetary surprises may convey private central bank information about the state of the economy.The event window starts 10-20 minutes before the press release and ends 10-20 minutes after the press conference.Following Jarociński and Karadi (2020), we further use sign restrictions to separate information effects from conventional monetary policy shocks.The identifying restriction is that monetary policy shocks should move interest rates and stock prices in opposite directions, while central bank information moves them in the same direction.
Our baseline shock series is constructed from high-frequency changes in the 6-months ahead OIS rate provided by Altavilla et al. (2019). 8While surprises in the 3-month rate become minuscule during the zero lower bound (ZLB) episode, we observe non-negligible surprises in the 6-month rate throughout our sample.We aggregate the daily shocks into quarterly frequency.Daily shocks are assigned fully to the current quarter if they occur on the Ąrst day of the quarter.If they occur within the quarter, they are partially assigned to the current and subsequent quarter (Gorodnichenko and Weber, 2016).The monetary policy shock series covers 1999Q1 through 2018Q4.Table 1 shows descriptive statistics and Figure A.1 in the Appendix shows the time series.
As a plausibility check and to provide a benchmark for our subsequent worker-level results, we estimate the responses of macroeconomic aggregates for the Austrian economy to the monetary policy shocks, see Figure A.2 in the Appendix.We Ąnd that a one-standard deviation monetary policy shock lowers real GDP by up to 0.4% with the peak effects attained between one and two years after the shock.We observe a similar dynamic for the employment rate which falls by up to 0.3 p.p. for prime-age workers.

Employment Probability
In this section, we estimate the effects of monetary policy shocks on the employment probability of workers.We Ąnd that low-paid workers who are employed in high-paying Ąrms before the shock are most affected by monetary policy.

Average Response
Before studying the distributional employment effects of monetary policy, we estimate the average employment effect across all workers.This provides a benchmark for the subsequent analysis.We estimate the following worker-level panel local projections on around 200 million worker-quarter observations of our baseline sample: for h = 0, . . ., 12 quarters, where e i,t+h denotes a binary employment variable with We include only workers in the regression that are employed in t − 1, the quarter preceding the monetary policy shock.This facilitates the comparison with the subsequent analysis, in which we need to condition on employment in t − 1 in order to study the responses by worker and Ąrm types.10On the right-hand side, α h i denotes a worker Ąxed effect (not the AKM worker Ąxed effect), ε MP t is the monetary policy shock, and Z i,t−1 is a vector of control variables, notably a linear time trend and season Ąxed effects for the four quarters.The coefficient of interest is β h , which captures the change in the employment probability in response to a monetary policy shock.Figure 1 shows the average response of the employment probability based on (3.1).The solid line shows the point estimates of β h , normalized to correspond to a one-standard deviation monetary policy shock, and the shaded areas indicate 68% and 95% conĄdence bands based on standard errors that are two-way clustered by worker and quarter.We Ąnd that the employment probability signiĄcantly falls.The response gradually builds up and peaks at a 0.27 p.p. lower employment probability Ąve quarters after the shock.The average workerlevel response is broadly in line with the aggregate employment response in Figure A.2.While Figure 1 shows the employment response of workers employed in the quarter before the monetary policy shock, we also examine the effect on workers who are unemployed before the shock.Figure A.3 in the Appendix shows that unemployed workers are signiĄcantly less likely to become employed after monetary policy shocks.In response to a one standard deviation shock, their employment probability falls by up to 0.89 p.p.In comparison, the average quarterly UE transition rate is 24.8% (see Table 1).

Heterogeneity across Worker and Firm Fixed Effects
We next present our empirical results on the distributional employment effects of monetary policy across worker and Ąrm Ąxed effects.Formally, we estimate the following statedependent worker-level panel local projections where β h captures the employment response of a worker with an average worker Ąxed effect in the year preceding the monetary policy shock (i.e., for W rolling i,τ −1 = W rolling i,τ −1 ) and an average Ąrm Ąxed effect for the Ąrm which employed the worker in quarter t − 1 (i.e., for F rolling j(i,t−1),τ −1 = F rolling j(i,t−1),τ −1 ).The coefficient γ W,h captures the differential employment response of a higher worker Ąxed effect, γ F,h captures the differential employment response of a higher Ąrm Ąxed effect, and γ W F,h captures the differential employment response of the interaction between a higher worker and a higher Ąrm Ąxed effect.11While we study the heterogeneity in our baseline with a linear speciĄcation, we show in the appendix (see Figure A.4) that our results are very similar if we use worker and Ąrm groups instead.Figure 2 presents our main results from equation (3.2).Panel (a) shows that workers with higher worker Ąxed effect are signiĄcantly less likely to become non-employed after a monetary policy shock (conditional on an average Ąrm Ąxed effect).The estimated differences are economically meaningful.Workers with a one standard deviation higher worker Ąxed Ąxed effect are up to 0.07 p.p. less likely to become non-employed compared to the average employment probability response of up to 0.27 p.p. Turning to the role of Ąrm Ąxed effects, panel (b) shows that workers employed in Ąrms with a higher Ąrm Ąxed effect are signiĄcantly more likely to become non-employed after a monetary policy shock (conditional on an average worker Ąxed effect).The magnitudes are similarily economically meaningful as for worker Ąxed effects.Equation (3.2) also contains an interaction effect between the worker shows the total employment response of different worker groups estimated based on β h , γ W,h , γ F,h , γ W F,h at h = 5 and the associated standard errors are in parantheses.For example, the employment response of high-paid workers in low-paying Ąrms is estimated based on , where p W x and p F x denote the x-th percentiles of the distribution of worker and Ąrm Ąxed effects, and σ W and σ F are the associated standard deviations.
and Ąrm Ąxed effects.Panel (c) shows that the coefficient on the interaction is signiĄcantly positive.This means that workers with combinations of high (or low) worker and Ąrm Ąxed effects are less likely to become non-employed than workers with opposite combinations.Put differently, workers are more likely to become non-employed when their worker Ąxed effect is in the opposite half of the distribution as their Ąrm Ąxed effect.Panel (d) of Figure 2 presents the group-speciĄc total employment responses, based on combining the average (β h ) and the differential (γ W,h , γ F,h , γ W F,h ) responses.We deĄne low and high-paid workers as workers with a worker Ąxed effect at the 25th and 75th percentile, respectively.Analogously, we deĄne low and high-paying Ąrms as Ąrm Ąxed effect at the 25th and 75th percentile across all workers, respectively.The table in panel (d) shows the employment response of different combinations of low and high-paid workers and low and high-paying Ąrms at horizon h = 5, when the average employment response peaks.We Ąnd that the employment responses differ similarly across Ąrm and worker types (see the ŞAllŤ column and row, respectively).While a monetary policy shock lowers the employment probability by 0.16 p.p. for workers at low-paying Ąrms, it plummets by 0.30 p.p. at highpaying Ąrms.In comparison, the drop is 0.23 p.p. for high-paid workers and 0.32 p.p. for low-paid workers across all Ąrms.What stands out from the table is that low-paid workers at high-paying Ąrms are most affected by monetary policy shocks.The employment probability for them drops by 0.36 p.p.The least affected group is high-paid workers from low-paying Ąrms, for which the employment probability drops by 0.15 p.p.This implies that the most affected group of workers in the table has a 2.4 times higher probability of non-employment than the least affected group.

Reallocation of Workers across Firms
In this section, we estimate the effects of monetary policy shocks on the reallocation of workers across Ąrms.We Ąnd that workers are more likely to switch Ąrms and they tend to switch to worse-paying Ąrms.In particular, low-paid workers employed by low-paying Ąrms before the shock are most likely to switch to worse-paying Ąrms.

Firm Switching Probability
To estimate the average effects of monetary policy shock on the probability that a worker switches between Ąrms, we use equation (3.1) but replace the left-hand side by a dummy variable that indicates whether a worker switches Ąrms 1 if a worker is employed in t + h by a different Ąrm than in t − 1, 0 else. (4.1) For h = 0, the sample average of e switch i,t+h is the quarterly Ąrm switching probability, the EE transition rate, which is 2.8% (see Table 1).The estimated average response of the Ąrm switching probability to a one standard deviation monetary policy shock is shown in Figure 3.The switching probability increases by up to 0.25 p.p. after the shock, which is a sizable increase over the average switching probability.However, the response is only mildly signiĄcant, in particular when compared to the response of the employment probability in Figure 1.We again turn to the question of which workers are more prone to change employers.In particular, we use (3.2) but replace again the left-hand side by the dummy variable indicating a change in employer from equation (4.1). Figure 4 provides our Ąndings.Most remarkable is the role of the worker Ąxed effect.Low-paid workers are signiĄcantly more likely to switch Ąrms.A one standard deviation lower worker Ąxed effect lowers the Ąrm switching probability by up to 0.12 p.p.In contrast, we donŠt Ąnd signiĄcant differences across Ąrm Ąxed effects or along the interaction of worker and Ąrm Ąxed effects.

Firm Wages
The previous section showed that monetary policy induces workers to switch employers, with the effect concentrated among low-paid workers.This naturally leads to the question where these worker move to, in particular, whether they Ąnd better-or worse-paying employers compared to before.Thus, we Ąrst ask whether monetary policy on average leads to a reallocation of workers towards lower or higher Ąrm Ąxed effects.To estimate the average effect of monetary policy shocks on the change in the Ąrm Ąxed effects of workers that switch given a one standard deviation above-average worker and Ąrm Ąxed effect.The inner and outer shaded areas respectively indicate 68% and 95% conĄdence bands two-way clustered by worker and quarter.Panel (d) shows the total Ąrm switching response of different worker groups estimated based on β h , γ W,h , γ F,h , γ W F,h at h = 5 and the associated standard errors are in parentheses.For example, the Ąrm switching response of high-paid workers in low-paying Ąrms is estimated based on x and p F x denote the x-th percentiles of the distribution of worker and Ąrm Ąxed effects, and σ W and σ F are the associated standard deviations.
Ąrms, we use (3.1) but replace the left-hand side by which is the change in the worker-associated Ąrm Ąxed effect between the original employer in t − 1 and the employer in t + h.Recall that in Section 3, we classiĄed workers and Ąrms using the backward-looking Ąxed effects in order to avoid endogeneity of Ąxed effects with respect to the monetary policy shocks.In contrast, (4.2) features the Ąrm Ąxed effect estimates over the entire sample, because we cannot otherwise compare Ąrm Ąxed effects over time.We estimate the regression on changes in the Ąrm Ąxed effect on the subset of workers switching Ąrms between period t − 1 and t + h. Figure 5 shows that the average response of the Ąrm Ąxed effect is signiĄcantly negative.After a one standard deviation monetary policy shock, the average change in the Ąrm wage premium of workers who switch Ąrms falls by up to 0.16%.These effects are sizeable, as compared to the unconditional average drop in the Ąrm Ąxed effect of 1.6% for switching workers.We next study the heterogeneity of the change in Ąrm Ąxed effects across workers and Ąrms.In particular, we use (3.2) but replace again the left-hand side by (4.2). Figure 6 provides our Ąndings.Panel (a) shows that the differential responses of changes in the Ąrm Ąxed effect associated with a higher worker Ąxed effect are indistinguishable from zero when the original Ąrm Ąxed effect equals the sample average.Similarly, panel (b) shows that the differential responses of changes in the Ąrm Ąxed effect associated with a higher Ąrm Ąxed effect are insigniĄcant when the worker Ąxed effect equals the sample average.Interestingly, panel (c) shows that there is a strong interaction between the worker Ąxed effect and the initial Ąrm Ąxed effect.Taking the average and all differential estimates together, panel (d) shows that and for a one standard deviation above-average worker and Ąrm Ąxed effect.The inner and outer shaded areas respectively indicate 68% and 95% conĄdence bands two-way clustered by worker and quarter.Panel (d) shows the total response of Ąrm Ąxed effects of different worker groups estimated based on β h , γ W,h , γ F,h , γ W F,h at h = 5 and the associated standard errors are in parentheses.For example, the Ąrm Ąxed effect response of high-paid workers in low-paying Ąrms is estimated based on , where p W x and p F x denote the x-th percentiles of the distribution of worker and Ąrm Ąxed effects, and σ W and σ F are the associated standard deviations.low-paid workers employed at low-paying Ąrms before the shock are losing the most from reallocation after monetary policy shocks.Overall, our results show that monetary policy shocks tends to reallocate workers toward worse-paying Ąrms.This effect is particularly pronounced for low-paid workers originally employed by low-paying Ąrms.

Sensitivity Analysis
In this section, we examine the sensitivity of our empirical Ąndings with respect to an alternative regression speciĄcation, alternative monetary policy shocks, control variables, sample, and data treatment.
Dummies for worker and Ąrm Ąxed effects groups.Our Ąndings on the role of worker and Ąrm Ąxed effects in Figures 2, 4, and 6 are estimated based on the local projection model in (3.2), which features linear interactions between monetary policy shocks and worker and Ąrm Ąxed effects.We examine the sensitivity of our Ąndings to an alternative semi-parametric regression model, in which we replace the linear interactions by dummies signifying whether worker and Ąrm Ąxed effects are above the average.Formally, we estimate e i,t+h (5.1) where ✶{•} is a binary dummy and Z i,t−1 is deĄned as in Section 3. To be precise, we estimate (5.1) when replacing the left-hand side by the Ąrm switching dummy in (4.1).Our Ąndings are similar to using the linear interactions (see panel (d) in Figure 4).The group with the highest exposure to monetary policy remain low-paid workers employed at low-paying Ąrms before the shock.Panel (c) of Figure A.4 shows the non-linear estimates of Ąrm Ąxed effect responses for workers switching Ąrms after the shock.To be precise, we estimate (5.1) when replacing the left hand side by the change in the Ąrm Ąxed effect in (4.2).Our Ąndings are overall robust to using the linear interactions, compare with panel (d) in Figures 6.The group with the highest exposure to monetary policy remain low-paid workers employed at low-paying Ąrms before the shock.
Monetary policy shocks.Our baseline monetary policy shocks are based on the signrestricted changes in the 6-month OIS rates.We examine the robustness of our results when using instead the changes in the 6-month OIS rates around policy announcement without applying sign restrictions.Figure A.5 shows that our estimated employment responses have similar point estimates, but are mostly insigniĄcant.This suggests that the raw surprises are strongly contaminated by information effects (Jarociński and Karadi, 2020).We further consider the sign-restricted 3-month OIS rate surprises.Figure A.6 shows that we obtain very similar effects to the baseline, both in terms of magnitude and signiĄcance.

Control variables.
We examine the sensitivity of our baseline speciĄcation to controlling for a set of standard macroeconomic variables.In particular, we enrich Z i,t−1 to include a lagged monetary policy shock and changes in log GDP, log CPI, and the employment rate.Missing worker observations.Our baseline data treatment only considers workers which are registered as employed or unemployed.Some workers leave our sample for some quarters before returning.Potential reasons are that they stopped receiving unemployment beneĄts, they left the country, or they became self-employed.We revisit our results when assuming that missing observations between two appearance of a worker in the sample are non-employment spells.Figure A.9 shows that this change ampliĄes the average employment response to -0.41 p.p. and increases heterogeneity in worker Ąxed effects.In contrast, Ąrm Ąxed effects become less important.

Conclusion
In this paper, we empirically characterize the distributional effects of ECB monetary policy shocks across workers and Ąrms using Austrian social security records.We focus on the heterogeneity across worker and Ąrm types identiĄed by a Abowd et al. (1999) regression, which is the workhorse model to estimate the worker and Ąrm components of wages.We document three novel results.First, we document which type of workers and Ąrms face the highest decline in employment in response to a contractionary monetary policy shock.Individuals who are low-paid and employed at high-paying Ąrms face the strongest employment declines.Second, monetary tightening increases the rate at which workers reallocate across Ąrms, in particular for low-paid workers.Third, we document that monetary policy shocks lead to a reallocation of workers to worse-paying Ąrms, with low-paid workers from low-paying Ąrms especially prone to falling off the Ąrm wage ladder.While all lowpaid workers are especially exposed to contractionary monetary policy shocks, we document large differences across low-paid workers depending on the type of Ąrm they are employed at before the shock.
Our results have implications for inequality, allocative efficiency, and transmission of monetary policy.For inequality, we show that the collapse of a job ladder is driven by the poorest workers.At the bottom of the income distribution, income is driven by labor earnings and its extensive margin (e.g., Amberg et al., 2022).Hence, the lower employment probabilities and the reallocation down a Ąrm wage ladder for the low-paid worker increases income inequality after a monetary shock.For allocative efficiency, if worker Ąxed effects correspond to workersŠ skills and productivity, and if the Ąrm Ąxed effects correspond to ĄrmsŠ productivity, reallocation towards lower-paying Ąrms could contribute to a drop in aggregate productivity, as is well-documented in the literature (e.g., Jordà et al., 2020;Meier and Reinelt, 2022;Baqaee et al., 2022).For the transmission of monetary policy, our results suggest that studying monetary models with two-sided heterogeneity is important.Moreover, our results suggest that a key moment is how the marginal propensity to consume is distributed across both worker and Ąrm types.

Figure 3 :
Figure 3: Average response of Ąrm switching probability

Figure 6 :
Figure 6: Firm Ąxed effect response across worker and (original) Ąrm Ąxed effects (a) Worker Ąxed effect (γ W,h ) Panel (a) ofFigure A.4  in the Appendix shows the group-speciĄc employment responses estimated from (5.1).Our Ąndings change little compared to using linear interactions (see panel (d) in Figure2).The estimated magnitudes are comparable and similarly signiĄcant.Importantly, the group with the highest non-employment exposure to monetary policy remain low-paid workers employed at high-paying Ąrms before the shock.Panel (b) of Figure A.4 in the Appendix shows the group-speciĄc Ąrm switching responses estimated from (5.1).Our Ąndings change little compared to using linear interactions (see panel (d) in Figures 4).The estimated magnitudes are comparable and similarly signiĄcant.Importantly, the group with the highest Ąrm switching exposure to monetary policy remain low-paid workers employed at high-paying Ąrms before the shock.Panel (b) of Figure A.4 shows the non-linear estimates of the group-speciĄc responses of the Ąrm switching probability.
Figure A.7  shows that this does not change our Ąndings much.Pre-ZLB sample.Every paper using high-frequency identiĄed monetary policy shocks faces the potential problem of the Zero Lower Bound (ZLB).Our baseline results use the longest possible sample including the ZLB.Importantly, because our monetary policy shocks are based on 6-month interest rates, we observe many shocks even during the ZLB episode (seeFigure A.1).Nevertheless, because monetary transmission may have changed we revisit our results in a pre-ZLB sample, ending in 2012Q2 just before the deposit facility rate reached zero. Figure A.8 in the Appendix shows that the employment responses are robust to using the pre-ZLB sample.

Figure
Figure A.2: Macroeconomic responses to monetary policy shocks (a) Gross Domestic Product where Z t−1 contains a linear time trend, one lag of the shock ε MP t and four lags of the employment rate, GDP growth, and CPI growth.The left hand side y t+h is ∆ h log GDP t+h in panel (a), ER t+h in panels (b)-(c), and ∆ h log CP I t+h in panel (d).The β h coefficients are standardized to capture the response to a one standard deviation increase in ε MP t .The inner and outer shaded areas respectively indicate 68% and 95% Newey-West conĄdence bands.

Figure
Figure A.3: Employment probability of initially unemployed workers

Figure
Figure A.4: Group-speciĄc responses using the non-linear speciĄcation (a) Response of employment probability

Figure A. 8 :
Figure A.8: Employment response for the pre-ZLB period (a) Average effect (β)

Figure
Figure A.9: Employment response when Ąlling missing observations (a) Average effect (β)