Creative Destruction and Subjective Well-Being

Abstract

In this paper we analyze the relationship between turnover-driven growth and subjective wellbeing. Our model of innovation-led growth and unemployment predicts that: (i) the effect of creative destruction on expected individual welfare should be unambiguously positive if we control for unemployment, less so if we do not; (ii) job creation has a positive and job destruction has a negative impact on wellbeing; (iii) job destruction has a less negative impact in US Metropolitan Statistical Areas (MSA) within states with more generous unemployment insurance policies; (iv) job creation has a more positive effect on individuals that are more forward-looking. The empirical analysis using cross-sectional MSA-level and individual-level data provide empirical support to these predictions.

1 Introduction

Does higher (per capita) GDP or GDP growth increase happiness? The existing empirical literature on happiness and income looks at how various measures of subjective wellbeing (SWB) relate to income or income growth, but without looking in further detail at what drives the growth process and at how the determinants of growth affect wellbeing. In this paper, we provide a first attempt at filling this gap.

More specifically, we look at how an important engine of growth, namely Schumpeterian creative destruction with its resulting flow of entry and exit of firms and jobs, affects SWB differently for different types of individuals and in different types of labor markets.

Thus, in the first part of the paper we develop a simple Schumpeterian model of growth and unemployment to organize our thoughts and generate predictions on the potential effects of turnover on life satisfaction. In this model, growth results from quality-improving innovations. Each time a new innovator enters a sector, the worker currently employed in that sector loses her job and the firm posts a new vacancy. Production in the sector resumes with the new technology only when the firm has found a new suitable worker. Life satisfaction is captured by the expected discounted valuation of an individual's future earnings. In the model a higher rate of turnover has both direct and indirect effects on life satisfaction. The direct effects are that, everything else equal, more turnover translates into both a higher probability of becoming unemployed for the employed, which reduces life satisfaction, and a higher probability for the unemployed to find a new job, which increases life satisfaction. The indirect effect is that a higher rate of turnover implies a higher growth externality and therefore a higher net present value of future earnings: this enhances life satisfaction. Overall, a first prediction of the model is that a higher turnover rate increases wellbeing more when controlling for aggregate unemployment than when not controlling for aggregate unemployment. A second prediction is that job creation increases and job destruction decreases wellbeing. A third prediction is that job destruction has a less negative effect on wellbeing, the more generous are unemployment benefits. A fourth prediction is that job creation increases future wellbeing more for more forward-looking individuals.¹

In the second part of the paper we test the predictions of the model using cross-sectional Metropolitan Statistical Area (MSA)-level US data. To measure creative destruction we follow Davis et al. (1996) and use their measure of job turnover, defined as the job creation rate plus the job destruction rate.² The data come from the Census' Business Dynamics Statistics (BDS) and are at the MSA level. In addition, we also use the Longitudinal Employer-Household Dynamics (LEHD) data from the Census, which provide information on hires, separations, employment, and thus turnover, also at the MSA level. To measure SWB, we use the Cantril ladder of life from the Gallup Healthways Wellbeing Index (Gallup), which asks individuals about both current and future wellbeing. The Cantril ladder is based on the following questions: “Imagine a ladder with steps numbered from 0 at the bottom to 10 at the top; the top of the ladder represents the best possible life for you and the bottom of the ladder represents the worst possible life for you. On which step of the ladder would you say you personally feel you stand at this time? And which level of the ladder do you anticipate to achieve in five years?”³

We investigate whether Schumpeterian creative destruction affects these measures of wellbeing positively or negatively, by regressing our measures of SWB on our creative destruction variables. The empirical analysis using cross-sectional MSA-level data on SWB and job turnover vindicates the theoretical predictions: namely, we find that: (i) the effect of creative destruction on wellbeing is positive when we control for MSA-level unemployment and less so if we do not; (ii) the effects of job creation and job destruction on wellbeing are positive and negative, respectively, and; (iii) job destruction has less negative effect when unemployment benefits are higher. Moreover, we find some evidence that job creation has a more positive impact on future wellbeing for more forward-looking individuals when we use income, age, and education to proxy for patience. These results are not only consistent with the theory, but they are also remarkably robust. In particular they hold whether looking at wellbeing at MSA level or at individual level, or whether using the BDS or the LEHD data to construct our proxy for creative destruction.

The paper relates to two main strands of literature. First, to the literature on innovation-led growth, job turnover and unemployment [e.g., see Davis et al. (1996), Mortensen and Pissarides (1998), Aghion and Howitt (1994, 1998), and Aghion et al. (2014)]. In particular Aghion and Howitt (1994, 1998) and Mortensen and Pissarides (1998) develop Schumpeterian models of growth through creative destruction, where growth is driven by quality-improving innovations by new entrants that make existing firms and jobs become obsolete. In any sector where new entry occurs, the incumbent firm closes down, therefore the worker employed by the incumbent firm loses her job whereas the entering firm in the sector posts a new vacancy. Equilibrium unemployment results from assuming labor market frictions in the form of a Poisson matching rate between new vacancies and workers looking for a new job. These two papers point to two opposite effects of growth on unemployment. One is a “capitalization” effect whereby more growth reduces the rate at which firms discount the future returns from creating a new vacancy: this effect pushes towards creating more vacancies and thus towards reducing the equilibrium unemployment. The counteracting effect is a “creative destruction” effect whereby more growth implies a higher rate of job destruction which in turn tends to increase the equilibrium level of unemployment. We contribute to this literature by looking at the counteracting effects of innovation-led growth on SWB.

Second, the paper contributes to the literature on SWB. In spite of a now large literature on self-reported wellbeing,⁴ there is no general consensus on how seriously these SWB measures should be taken, or on exactly what they mean. Indeed some of the most exciting recent work [e.g., see Benjamin et al. (2012, 2014)] is investigating these fundamental questions.⁵ In this paper, we find that life satisfaction responds to the future growth prospects that are inherent in creative destruction, even in spite of the related short-run unemployment effects, and at the same time we provide some evidence of the validity and usefulness of self-reported wellbeing as a measure of expected future material wellbeing. Such findings have not previously been documented in the wellbeing literature and they provide further evidence of the usefulness of these wellbeing measures.

The paper is organized as follows. Section 2 develops the model and generates predictions on the effects of turnover on SWB, and how these effects depend upon individual or local labor market characteristics. Section 3 describes the data, the approach underlying the empirical analysis, and presents the empirical results. Section 4 considers several robustness checks. Section 5 concludes the paper. The Appendix contains the proofs and the Online Appendix presents extensions to the baseline framework, additional proofs, and extra tables.

2 Theoretical Analysis

2.1 A Toy Model

In this section, we offer a simple model to motivate our empirical analysis. The source of economic growth is Schumpeterian creative destruction which at the same time generates endogenous obsolescence of firms and jobs. The workers in the obsolete firms join the unemployment pool until they are matched to a new firm. Higher firm turnover has both a positive effect (by increasing economic growth and by increasing employment prospects of unemployed workers) and a negative effect (by increasing the probability of currently employed workers losing their job) on wellbeing. Which effect dominates will in turn depend upon both individual characteristics (e.g., discount rate and risk-aversion) and characteristics of the labor market (e.g., unemployment benefits). To keep the analysis tractable, in what follows we will consider a steady-state economy with exogenous entry, risk-neutral agents, and only endogenous job destruction. These assumptions will be relaxed in the Online Appendix: Section A1.1 focuses on transitional dynamics, A1.2 considers a model with exogenous job destruction, A1.3 considers the implications of risk aversion, and A1.4 endogenizes entry in the theoretical model.

2.1.1 Production Technology and Innovation

We consider a multi-sector Schumpeterian growth model in continuous time. The economy is populated by infinitely-lived and risk-neutral individuals of measure one, and they discount the future at rate ρ. Therefore the household Euler equation is simply

$$r=\rho ,$$ (1)

where r is the interest rate of the economy.

The final good is produced using a continuum of intermediate inputs, according to the logarithmic production function:

$$ln{Y}_t={\int }_{j\in \mathcal{J}}ln{y}_{jt}dj$$

where 𝒥 ⊂ [0, 1] is the set of active product lines. We will denote its measure by J ∈ [0, 1]. The measure J is invariant in steady state.

Each intermediate firm produces using one unit of labor according to the following linear production function,

$$y_{jt}=A_{jt}{l}_{jt},$$

where l_jt = 1 is the labor employed by the firm, and is the same in all sectors. Thus the measure of inactive product lines is equal to the unemployment rate

$$u_t=1-J_t,$$

where u denotes the equilibrium unemployment rate. Our focus will be on balanced growth path equilibrium, therefore, when possible, we will drop time subscripts to save notation.

2.1.2 Innovation and Growth

An innovator in sector j at date t will move productivity in sector j from A_jt−1 to

$$A_{jt}=\lambda A_{jt-1},$$

where λ > 1. The innovator is a new entrant, and entry occurs in each sector with Poisson arrival rate x which we assume to be exogenous. Upon entry in any sector, the previous incumbent firm becomes obsolete and its worker loses her job and the entering firm posts a new vacancy with an instantaneous cost cY.⁶ Production in that sector resumes with the new technology when the firm has found a new suitable worker.

2.1.3 Labor Market and Job Matching

Following Pissarides (1990), we let

$$m(u_t,v_t)=u_t^{\alpha }{v}_t^{1-\alpha }$$ (2)

denote the arrival rate of new matches between firms and workers, where u_t denotes the number of unemployed at time t and v_t denotes the number of vacancies. Thus the flow probability for each unemployed worker to find a suitable firm is

$$m(u_t,v_t)/u_t,$$

whereas the probability for any new entrant firm to find a suitable new worker is

$$m(u_t,v_t)/v_t.$$

In steady state, there will be a constant fraction of product lines that are vacant (of measure v), and the remaining fraction will be producing. We illustrate this economy in Figure 1:

Figure 1. Model Economy.

Finally, we assume that in each intermediate sector where a worker is currently employed, the worker appropriates fraction β of profits whereas the complementary fraction (1 − β) accrues to the employer.

2.1.4 Valuations and Life Satisfaction

Life satisfaction is captured by the average present value of an individual employee, namely:

$$W_t=u_t{U}_t+(1-u_t)E_t,$$

where U_t is the net present value of an individual who is currently unemployed, and E_t is the net present value of an individual who is currently employed.

The value of being currently employed, satisfies the asset equation:

$$\rho E_t-{\dot{E}}_t=w_t+x(U_t-E_t).$$

In words: the annuity value of being currently employed is equal to the capital gain Ė_t plus the wage rate w_t at time t and with arrival rate x the worker becomes unemployed as the incumbent firm is being displaced by a new entrant. Here we already see the negative effect of turnover on currently employed workers.

Similarly, the value of being unemployed satisfies the asset equation:

$$\rho U_t-{\dot{U}}_t=b_t+(m(u_t,v_t)/u_t)(E_t-U_t).$$

As before, the annuity value of being currently unemployed is equal to the capital gain U̇_t plus the benefit b_t accruing to an unemployed worker,⁷ and with arrival rate m(u_t, v_t)/u_t the unemployed worker escapes unemployment. For any given unemployment rate, turnover has a positive effect on the value of being unemployed because it creates job opportunities.

2.2 Solving the Model

We now proceed to solve the model for equilibrium production and profits, the equilibrium steady-state unemployment rate, the steady-state growth rate, and the equilibrium value of life satisfaction.

2.2.1 Static Production Decision and Equilibrium Profits

Let w_t denote the wage rate at date t. The logarithmic technology for final good production implies that the final good producer spends the same amount Y_t on each variety j. As a result, the final good production function generates a unit elastic demand with respect to each variety: y_jt = Y_t/p_jt.

Note that the cost of production is simply w_jt which is the firm-specific wage rate. Then the profit is simply

$${\pi }_{jt}=p_{jt}{y}_{jt}-w_{jt}=Y_t-w_{jt}.$$ (3)

Next, the above sharing rule between wage and profits implies that w_jt = β(Y_t − w_jt), hence

$$w_{jt}=w_t=\frac{\beta }{1+\beta }{Y}_t,\text{and}{\pi }_{jt}=\frac{1}{1+\beta }{Y}_t=\pi Y.$$

Clearly, β determines the allocation of income in the economy, with a higher β shifting the income distribution towards workers.

2.2.2 Steady-state Equilibrium Unemployment

Our focus is on a steady-state equilibrium in which all aggregate variables (Y_t, w_t, U_t, E_t) grow at the same constant rate g, and where the measure of unemployed u and the number of vacancies and the interest rate remains constant over time.⁸ Henceforth, we will drop the time index t, when it causes no confusion.

In steady state, the flow out of unemployment must equal the flow into unemployment. Namely:

$$m(u,v)=(1-u)x.$$ (4)

The left-hand side is the flow out of unemployment, the right-hand side is the flow into unemployment, equal to the number of active sectors (1 − u) time the turnover rate x.

In addition, the number of sectors without an employed worker is equal to the number of sectors with an open vacancy, u = v. Combining this fact with the matching technology (2), we get:

$$m=u=v.$$ (5)

Putting equations (4) and (5) together, we obtain the equilibrium unemployment rate u = (1 − u)x, or equivalently

$$u=\frac{x}{1+x}.$$ (6)

That the numerator of u is increasing in x reflects the job destruction effect of turnover on currently employed workers; that the denominator is also increasing in x reflects the positive effect a higher turnover rate has on the job finding rate of currently unemployment workers.

The first effect dominates here, with the equilibrium unemployment rate increasing in the turnover rate x. However this very much hinges on the fact that innovative turnover is the only source of job destruction in this baseline model. In Online Appendix A1.2, we introduce the possibility of exogenous job destruction on top of innovation-driven job destruction. Then we show that the higher the exogenous rate of job destruction, the more the innovation rate x contributes to reducing unemployment, and therefore the more positive the overall effect of x on equilibrium wellbeing.

Now we can express the growth rate of the economy.

Lemma 1. The balanced growth path growth rate of the economy is equal to

$$g=mln\lambda ,$$

where m denotes the flow of sectors in which a new innovation is being implemented (i.e., the rate at which new firm-worker matches occur).

Proof. See Appendix A.

Then, using the fact that in steady-state equilibrium we have $m=u=\frac{x}{1+x}$, we get the equilibrium growth rate as,

$$g=\frac{x}{1+x}ln\lambda .$$ (7)

As expected, the growth rate is increasing in the turnover rate x and with the innovation step size λ.

2.2.3 Equilibrium Valuations and Life Satisfaction

Recall that life satisfaction is the average welfare of an individual employee, namely:⁹

$$W=uU+(1-u)E,$$

where:

$$rE-\dot{E}=\beta \pi Y+x(U-E),\text{and}$$ (8)

$$rU-\dot{U}=bY+(m(u,v)/u)(E-U).$$ (9)

Now, after substituting for E and U in the expression for the steady-state value of W, and using the fact that in steady state Ė = gE and U̇ = gU, and that in equilibrium [see equation (5)] m = u = x/(1 + x), we get the following expression for life satisfaction:¹⁰

$$W=\frac{Y}{r-g}[\beta \pi -\frac{xB}{1+x}]$$ (10)

where

$$g=\frac{x}{1+x}ln\lambda \text{and}B=\beta \pi -b.$$ (11)

From the above expression for W, we see three effects of turnover on life satisfaction. First, for given growth rate g, more turnover increases the probability of an employed worker losing her current job (numerator in $\frac{xB}{1+x}$) which reduces life satisfaction; second, for given growth rate g, more turnover increases the probability of an unemployed worker finding a new job (denominator in $\frac{xB}{1+x}$) which increases life satisfaction; third, higher turnover increases the growth rate g which in turns acts favorably on life satisfaction: this is the capitalization effect mentioned in the introduction. The overall effect of turnover on life satisfaction is ambiguous.¹¹

2.2.4 Comparative Statics and Additional Discussions

In this section, we discuss the implications of our model.

Unemployment vs Capitalization Effect

If we look at the effect of turnover on life satisfaction controlling for unemployment, this effect is unambiguously positive. To see this, after some straightforward algebra we reexpress equilibrium wellbeing W as:

$$W=\frac{Y}{r-g}[ub+(1-u)\beta \pi ]$$ (12)

which for given u is increasing in x since it is increasing in g and g is increasing in x (capitalization effect).¹²

Taking the derivative of (10) with respect to x and substituting (11) we get:

$$\frac{\partial W}{\partial x}=\frac{Y[\beta \pi ln\lambda -B\rho ]}{{[(1+x)(\rho -ln\lambda )+ln\lambda ]}^2},$$

which is clearly positive when $\rho <\frac{\beta \pi ln\lambda }{B}$, i.e., when capitalization effect dominates the negative unemployment effect. Note also that life satisfaction increases more with turnover x the more generous unemployment benefits are:¹³

$$\frac{{\partial }^{2}W}{\partial x\partial b}>0.$$

We summarize the above discussion in the following proposition:

Proposition 1. (i) A higher turnover rate x increases life satisfaction W unambiguously once we control for the unemployment rate, not otherwise; (ii) life satisfaction increases more with turnover x the more generous unemployment benefits are.

Job Creation vs Job Destruction

So far, we have proxied job turnover using a single parameter x. However, we can also write (12) in terms of job creation and job destruction rates. Note that in our model, job creation happens through new matches, which happen at the rate m (= job_creation) and job destruction happens as incumbent firms are replaced by new entrants at the rate x (= job_destruction). Hence we can express (12) as

$$W=\frac{Y}{\rho -ln\lambda \times \mathit{\text{job}}\_\mathit{\text{creation}}}[\beta \pi -B\frac{\mathit{\text{job}}\_\mathit{\text{destruction}}}{1+\mathit{\text{job}}\_\mathit{\text{destruction}}}]$$ (13)

Clearly, we obtain the following immediate comparative statics:

$$\frac{\partial W}{\partial \mathit{\text{job}}\_\mathit{\text{creation}}}>0,\frac{\partial W}{\partial \mathit{\text{job}}\_\mathit{\text{destruction}}}<0,\text{and}\frac{{\partial }^{2}W}{\partial \mathit{\text{job}}\_\mathit{\text{destruction}}\partial b}>0.$$

Next, the following proposition follows.

Proposition 2. (i) A higher job creation rate increases, whereas a higher job destruction rate decreases life satisfaction W; (ii) life satisfaction decreases less with job destruction the more generous the unemployment benefits.

Current vs Future Wellbeing and Transitional Dynamics

In this section, we discuss briefly the transitional dynamics and its impact on wellbeing. Consider a sudden unexpected increase in the rate of creative destruction from x^old to x^new such that x^new > x^old. This generates a transition from the old steady state to a new steady state. During this transition, the path of the growth rate is summarized in the following lemma.

Lemma 2. Consider an initial steady state with a creative destruction rate of x^old. Assume that at time t = 0, the creative destruction rate becomes x^new. Then the growth rate during the transition can be expressed as

$$g_t=g_{ss}^{\mathit{\text{new}}}-e^{-t{x}^{\mathit{\text{new}}}}[g_{ss}^{\mathit{\text{new}}}-g_{ss}^{\mathit{\text{old}}}]$$ (14)

where g^new and g^old are the new and old steady-state growth rates, respectively and g^new > g^old.

Proof. See Online Appendix A1.1.

Let us denote the wellbeing at time t under x^old and x^new by $W_t^{\mathit{\text{old}}}$ and $W_t^{\mathit{\text{new}}}$, respectively. Moreover, let us denote their respective steady-state trajectories by $W_{t,ss}^{\mathit{\text{old}}}$ and $W_{t,ss}^{\mathit{\text{new}}}$. Expression (14) makes it clear that when there is an increase in creative destruction from x^old to x^new > x^old, the growth rate will monotonically converge toward its new level. The impact of this change is illustrated in Figure 2. Before time 0, i.e., at t < 0, wellbeing is increasing on its trajectory $W_{t,ss}^{\mathit{\text{old}}}$. When the turnover rate increases from x^old to x^new when t = 0, wellbeing accelerates and starts to evolve toward its new trajectory $W_t^{\mathit{\text{new}}}\to W_{t,ss}^{\mathit{\text{new}}}$.

Figure 2. Wellbeing during Transition.

The important point to note here is that the gap between the new trajectory and the old trajectory widens over time. For instance, the gap between $W_t^{\mathit{\text{new}}}-W_{t,ss}^{\mathit{\text{old}}}$ at time t = T₁ is smaller than the gap at time t = T₂. This implies that any change in the turnover rate has a bigger impact in the future wellbeing than the current wellbeing. Hence $\mathrm{\Delta }{W}_t=W_t^{\mathit{\text{new}}}-W_{t,ss}^{\mathit{\text{old}}}$ is increasing over time. In this economy, a given individual's expected period-T future wellbeing from a time-zero perspective can be expressed as

$$\mathit{\text{future}}\_\mathit{\text{wellbeing}}(T)=e^{-\rho T}{W}_T.$$

Clearly, an increase in turnover that increases future wellbeing will be perceived more highly by more patient individuals (with lower discount rate ρ). In Online Appendix A1.1, we show that the transition in our model happens very fast. Motivated by this fact, for any given wellbeing path W_t, if we approximate it with its steady-state value W_T ≈ W_ss, we can also show this formally:

$$\begin{array}{c}\frac{\partial \mathit{\text{future}}\_\mathit{\text{wellbeing}}(T)}{\partial x}\approx e^{-\rho T}\frac{\partial W_{ss}}{\partial x}>0,\text{and}\\ \frac{{\partial }^2\mathit{\text{future}}\_\mathit{\text{wellbeing}}(T)}{\partial x\partial \rho }\approx -T{e}^{-\rho T}\frac{\partial W_{ss}}{\partial x}+e^{-\rho T}\frac{{\partial }^2{W}_{ss}}{\partial x\partial \rho }<0.\end{array}$$

In words, future wellbeing increases in creative destruction, and more so for more patient individuals. A nice feature of our wellbeing data is that individuals are asked about their expectation about their future wellbeing as well. This will allow us to directly test this prediction of our model using the “future wellbeing” measure.

2.2.5 Summary and Main Predictions

In the empirical analysis below, we will use cross-MSA data on wellbeing and job turnover to test the following predictions from the model:

• Prediction 1: A higher turnover rate increases wellbeing more when controlling for aggregate unemployment than when not controlling for aggregate unemployment.

• Prediction 2: A higher job creation rate increases wellbeing, whereas a higher job destruction rate decreases wellbeing.

• Prediction 3: A higher turnover rate increases wellbeing more, whereas a higher job destruction rate decreases wellbeing less the more generous the unemployment benefits.

• Prediction 4: A higher turnover rate increases future wellbeing more for more forward-looking individuals.

3 Empirical Analysis

3.1 Data

The data on creative destruction come from the BDS, which provide, at the metropolitan level (MSA), information on job creation and destruction rates. The job creation (destruction) rate is the sum of all employment gains (losses) from expanding (contracting) establishments from year t–1 to year t including establishment creations (destructions), divided by the average employment between years t and t – 1. These rates are computed from the whole universe of firms as described in the Census Longitudinal Business Database. Our main measure of creative destruction is the “job turnover rate”, defined as the sum of the job creation and job destruction rates. We also analyze the role of creation rates and destruction rates separately. For our panel analysis, we use an alternative data source, the LEHD constructed by the Census bureau. This dataset varies at the quarterly level, whereas the BDS data vary only at the annual level. The LEHD dataset also allows for a sectoral breakdown, which we take advantage of to construct a predicted Bartik-like measure of turnover that we use as a robustness check. The job creation rate in the LEHD is defined as the estimated number of workers who start a new job in a given quarter divided by the average employment in that quarter. The job destruction rate is defined as the estimated number of workers whose job ended in a given quarter divided by the average employment in that quarter.

The main data source on SWB is the Gallup Healthways Wellbeing Index, which collects data on 1,000 randomly selected Americans each day through phone interviews. The period covered is 2008-2011. Subjective wellbeing in Gallup is assessed through various questions aimed at capturing different dimensions of wellbeing. We focus on the “Cantril ladder of life” questions which are intended to measure the individual's evaluation of her life. Each individual is asked: “Please imagine a ladder with steps numbered from 0 at the bottom to 10 at the top; the top of the ladder represents the best possible life for you and the bottom of the ladder represents the worst possible life for you; on which step of the ladder would you say you personally feel you stand at this time?”; and then “which level of the ladder do you anticipate to achieve in five years?”. We refer to answers to the first question as the “current ladder” and to the second one as the “future ladder”. The distinction between current and future ladder measures is particularly interesting, as we recall that some of the predictions, especially Prediction 4, rely mainly on future wellbeing.

To test the robustness of our main results to an alternative measure of wellbeing, we use the life satisfaction measure from the Behavioral Risk Factor and Surveillance System (BRFSS).¹⁴ The sample size is roughly similar to that of Gallup but the BRFSS does not distinguish between current and future wellbeing. Further details on these data are provided in the robustness section.

Additional data sources used are: the Local Area Unemployment Statistics from the Bureau of Labor Statistics for the MSA-level unemployment rate; the FBI Crime Statistics for the MSA crime rates; the Bureau of Economic Analysis for population levels; and the Department of Labor for states' unemployment insurance policies.

The descriptive statistics of our main data can be found in Table 1.

Table 1. Summary Statistics.

	Observations	Mean	Standard deviation	Min	Max
MSA-level 2008-2011 Averages (Used in Panel A of Table 2, 3, and 4)
Current ladder	363	6.724	0.192	6.059	7.431
Future ladder	363	7.950	0.195	7.363	8.571
Job turnover rate	363	0.261	0.036	0.165	0.409
Job creation rate	363	0.125	0.018	0.082	0.215
Job destruction rate	363	0.136	0.021	0.082	0.225
Unemployment rate	363	0.083	0.025	0.035	0.275
Log of income	363	8.127	0.164	7.517	8.605
Share African Americans	363	0.102	0.102	0	0.454
Population (in thousands)	355	726.5	1,621	55.24	19,533
Crime rate(/100,000 inhab.)	352	401.4	176.9	65.33	1085
Unemp. insurance generosity	363	1.759	0.520	0.624	2.931
Individual-level Data, 2008-2011(Used in Panel B of Table 2, 3, and 4, and Table A11)
Current ladder	556,719	6.722	1.950	0	10
Future ladder	544,620	8.032	1.983	0	10
Female	556,719	0.491	0.500	0	1
Age	556,719	39.83	11.91	18	60
Black	556,719	0.126	0.332	0	1
Asian	556,719	0.026	0.158	0	1
White	556,719	0.725	0.447	0	1
Married or living with partner	556,719	0.592	0.492	0	1
Average years of schooling	556,719	14.18	2.346	10	18
Monthly household income	556,719	5,709	5,033	347.1	16,483
Log income	556,719	8.234	0.984	5.850	9.710
Quarterly Panel Data (Used in Panel C of Table 2 and 3)
Job turnover rate	5,704	0.326	0.093	0.141	1.815
Job creation rate	5,704	0.146	0.045	0.053	0.894
Job destruction rate	5,704	0.180	0.053	0.080	0.929

3.2 Estimation Framework

Our measure of creative destruction varies at the MSA level, thus we estimate MSA-level regressions. However, in order to take advantage of our micro-level data on SWB, we also perform individual-level regressions that allow us to have a richer and more meaningful set of controls. Individual characteristics such as marital status do not vary much if we aggregate them at the MSA level, yet they are important determinants of wellbeing at the individual level. In all cases, regressions are OLS. We restrict the analysis to working age individuals (18-60 years old) to be closer to the model in which individuals are either employed or unemployed.^15,16

At the MSA level, we look at purely cross-sectional regressions, where we average our SWB data at the MSA level and across our sample years.¹⁷ In all specifications we control for MSA-level averages of the Gallup respondents' income. Income is measured in terms of household income brackets. We calculate the midpoints of these brackets assuming that income is log-normally distributed and we then average at the MSA level these log midpoints.¹⁸ In our regressions, we also explore what happens when we add MSA-level potential confounders such as crime rate, the share of African Americans, and population.

At the individual level, we perform regressions where we control for individual characteristics such as education, income, and ethnicity, as well as gender, marital status, and age. Our specification is as follows:

$${\mathit{\text{SWB}}}_{i,m,t}=\alpha \times X_{m,t}+\beta \times Y_{m,t}+\delta \times Z_{i,t}+T_t+{\mathrm{\epsilon }}_{i,t},$$ (15)

where SWB_m,t is SWB for individual i who lives in MSA m in year t. This measure is derived through the current ladder question or the future ladder question in the Gallup survey. X_m,t is either the job turnover rate and the unemployment rate in MSA m in year t (Prediction 1), or the job creation and the job destruction rates introduced separately (Prediction 2). Y_m,t are MSA-level controls, such as the population level in year t, the crime rate, and the share of African-Americans. Z_i,t are individual-level controls: gender, age, age square, 4 race dummies, 6 education dummies, 6 family status dummies, and 9 dummies for income brackets. T_t are year and month fixed effects. Finally, ε_m,t is the error term. A constant is also included and standard errors are clustered at the MSA level. When testing Prediction 3, we interact our creative destruction proxies with a measure of the generosity of the state's unemployment insurance. When testing Prediction 4, we interact the job creation and the job destruction rates with proxies for the individual discount rate (age, education, and income). Robustness checks are discussed below in Section 4.2.

3.3 Testing Prediction 1

In this section, we test Prediction 1: A higher turnover rate increases wellbeing more when controlling for aggregate unemployment than when not controlling for aggregate unemployment.

Recall that the model highlights two opposite forces whereby creative destruction impacts SWB: the negative effect that comes from the higher risk of unemployment through job destruction and the positive growth effect through new job creation. Controlling for the unemployment rate should capture part of the negative force of creative destruction and thus lead to a more positive coefficient of creative destruction on wellbeing than without the control for unemployment.

MSA-level Results

Before displaying the regression results, we show two scattered plots where one observation corresponds to an MSA. Figure 3 plots the MSA's average life satisfaction on MSA-level job turnover. We then regress these MSA-level life satisfaction and job turnover variables on the MSA's unemployment rate and plot the residuals in Figure 4. We see that wellbeing is more strongly positively associated with job turnover, once we control for the unemployment rate.

Figure 3. Simple Scatter Plot.

Figure 4. Residual Scatter Plot.

Now moving to the regression results, Table 2 Panel A shows the results from baseline OLS regressions at the MSA level. We see that the job turnover has a positive but statistically insignificant icant on current wellbeing. Column 2 shows that once we control for unemployment, job turnover has an effect on wellbeing that is more than twice as large and strongly statistically significant. This is in line with our model which predicts that, controlling for unemployment, turnover should have a more positive effect on wellbeing as it implies higher growth and a higher probability for currently unemployed workers of finding a new job.¹⁹ In Column 3, we add additional MSA controls: population, crime rate, and share of African Americans.²⁰ We see that these potential confounders do not significantly affect the coefficients of the creative destruction variable.

Table 2. Test of prediction 1.

VARIABLES	(1)	(2)	(3)	(4)
	Current ladder		Future ladder
Panel A: MSA-level Analysis
Job turnover rate	0.599 (0.361)	1.288 (0.410)	1.322 (0.424)	1.726 (0.306)
Unemployment rate		-2.727 (0.786)	-2.581 (0.823)	-0.930 (0.507)
Log of income	0.342 (0.0839)	0.195 (0.088)	0.225 (0.106)	0.297 (0.079)
Additional MSA controls			x	x
Observations	363	363	344	344
R-squared	0.100	0.198	0.217	0.459
p-value Job Turnover [1]= Job Turnover [2]		0.000
Panel B: Individual-level Analysis
Job turnover rate	0.0676 (0.236)	0.521 (0.237)	0.611 (0.285)	0.984 (0.148)
Unemployment rate		-2.299 (0.443)	-2.168 (0.502)	-0.0857 (0.298)
MSA-level log of income	-0.187 (0.048)	-0.285 (0.046)	-0.263 (0.051)	-0.0424 (0.038)
Additional MSA controls			x	x
Individual controls (incl. income)	x	x	x	x
Year and Month F.E.	x	x	x	x
Observations	556,300	556,300	461,054	450,908
R-squared	0.103	0.104	0.103	0.095
p-value Job Turnover [1]= Job Turnover [2]		0.000
Panel C: Panel Analysis
Job turnover rate	0.678	0.787	0.249	0.234
	(0.0970)	(0.105)	(0.142)	(0.140)
Unemployment rate		-2.238 (0.301)	-1.743 (1.054)	-1.074 (1.057)
Log of income	0.410 (0.0301)	0.360 (0.0307)	0.412 (0.0373)	0.192 (0.0390)
MSA F.E.			x	x
Additional MSA controls	x	x	x	x
Year and quarter F.E	x	x	x	x
Observations	4,884	4,884	4,884	4,884
R-squared	0.189	0.203	0.325	0.256
p-value Job Turnover [1]= Job Turnover [2]		0.041

The dependent variables are SWB measures from Gallup: columns (1) through (3) use the current Cantril ladder of life whereas column (4) uses the future ladder of life. Sample years are 2008-2011 and the sample is restricted to working-age respondents.

Column (1) regresses SWB on the job turnover rate, which comes in Panel A and B, from the BDS, and in panel C, from the LEHD which provide data at the quarterly level. Column (2) adds a control for the MSA-level unemployment rate. Columns (3) and (4) further add some MSA-level controls: population (in levels), crime rates, and share of African-Americans. All specifications control for income at the MSA-level.

All data sources and variables definitions are in the main text.

Panel A carries a cross-sectional analysis at the MSA-level, where the variables are averaged across years within each MSA. The SWB measures are averaged using the weights attached by Gallup to each respondent.

Panel B carries a repeated cross-section analysis at the individual-level. Month and Year fixed effects are added to the specification as well as individual controls: age, age square, a dummy for being female, 6 dummies for family status, 6 dummies for education, and 4 race dummies (black, asian, white, other or missing), as well as 9 dummies for household income brackets. Regressions are weighted by individual weights attached by Gallup to each respondent.

Panel C shows the results of a quarterly panel analysis at the MSA-level. The SWB measures are averaged at the MSA-quarter level using the weights attached by Gallup to each respondent. All regressions include year and quarter fixed effects, quarterly controls for the MSA's average log of income, and the share of African Americans in the MSA, as well as annual controls for population levels and crime rates. Columns (3) and (4) add MSA fixed effects.

The difference between the estimates on creative destruction in Column 1 versus 2 and 3 is statistically significant at the 1% level. The last row in each Panel of Table 2 reports the p-value associated with a Wald test of the hypothesis that α_CD;column1 = α_CD;column2.

Column 4 repeats the same specification as column 3 but with future wellbeing as a dependent variable. We first see that job turnover has a stronger effect on the future ladder than on the current Cantril ladder. This in turn points to the notion that individuals disentangle the short-run losses from becoming unemployed as a result of job turnover from the long-term gains associated with higher growth and more new job opportunities in the future.

The magnitude of the effect of creative destruction on current life satisfaction is in the same ballpark as that of the effect of the unemployment rate. In particular, moving from an MSA which is at the 25th percentile in terms of its level of creative destruction (i.e., with a job creation rate + destruction rate at 23.5%) to an MSA at the 75th percentile (i.e., with a job creation rate plus job destruction rate at 28.3%) is associated with an increase in the current ladder of life of 0.06 points (Column 2 in Table 2, Panel A). As a benchmark, looking at the same regression, moving from the 75th to the 25th percentile in terms of the unemployment rate (that is, from a 9.4% to a 6.7% unemployment rate) is associated with an increase in life satisfaction of 0.07 points. Another way to put it is that a one standard deviation increase in job turnover increases the current ladder by 0.25 standard deviation: that effect is equivalent to a 0.7 (= (0.036 × 1.288/2.727)/0.025) standard deviation decrease in the MSA-level unemployment rate.

Individual-level Results

In Table 2 Panel B we perform individual-level regressions and find qualitatively similar results as in Panel A. The difference is that we now also control for individual-level characteristics and for year and month fixed effects. We thus control for household income brackets and we keep the control for the MSA-level log of income. Note that the MSA-level income has a negative impact on wellbeing once we control for individual-level income, which suggests that wellbeing depends on relative income of an individual. The creative destruction variable now varies at the MSA-year level.

We can still reject at the 1% level that the coefficient of job turnover on wellbeing is the same whether or not we control for the unemployment rate. We also see that the effect of turnover is stronger on the future ladder than on the current ladder.

The magnitude of the creative destruction effect is smaller to that displayed at the MSA level. A one standard deviation increase in job turnover has an effect on the current ladder of life which is equivalent to a 0.3 standard deviation increase in the MSA-level unemployment rate (= (0.036 × 0.521/2.299)/0.025).

Panel Results

In Panel C, we show results from quarterly panel regressions with year and quarter fixed effects, with and without MSA fixed effects. However, we want to stress why we have a preference for the cross-sectional analysis. The theoretical concept of creative destruction is being proxied in our empirical analysis by a job turnover variable. So we are proxying x(t) by x*(t) which is equal to x(t) plus some measurement (or proxy) error ε(t). Adding MSA fixed effects into the regression changes in an unfavorable way the relative variances of the signal, the variance of x(t), and the noise, the variance of ε(t). If the job destruction variable changes only slowly over time within each MSA, which is the case here, looking at the deviation of job destruction from its MSA time-mean is going to be problematic, as more of that deviation is going to come from the proxy error, not from the variable itself. Hence our predictions are better captured by cross-sectional regressions than by panel regressions that cover such short time periods.

Because our sample period is very short, we use a quarterly frequency to look at panel specifications. Thus we use the LEHD dataset constructed by the Census Bureau, based on the Quarterly Census of Employment and Wages and other administrative and survey data. Indeed these data contain information on employment, earnings, and job flows at the MSA and quarterly level. In terms of creative destruction: rather than job creations and destructions, the data give us the number of hires and separations. To compute the turnover rates, we divide these hires or separations by the average stock of employment in that quarter. The results are reported in Table 2 Panel C.

If we compare column 1 to column 2, we see again that the coefficient for job turnover is higher when we control for the MSA-level unemployment rate than when we do not. The difference is significant at the 5% level. These two columns are without MSA fixed effects. When we add MSA fixed effects (Column 3), the coefficient of job turnover is still significantly positive although of a smaller magnitude. Note that all the specifications in Panel C control for time-varying potential MSA-level controls: population levels, crime rates, share of African Americans.

3.4 Testing Prediction 2

In this section, we test Prediction 2: A higher job creation rate increases wellbeing, whereas a higher job destruction rate decreases wellbeing.

MSA-level Results

Table 3 Panel A shows the results from the baseline OLS regressions at the MSA level. The first two columns use current ladder whereas the last two columns use the future ladder as dependent variables. In the first and third columns, we see the positive effect of job creation and the negative effect of job destruction on current and future wellbeing which are very much in line with Prediction 2. Columns 2 and 4 introduce additional MSA-level confounders: the MSA's average population level over the period, its average crime rate and its average share of African Americans. These controls do not change the pattern: overall, job creation and destruction have opposite effects on wellbeing, as the theory predicted.

Table 3. Test of prediction 2.

VARIABLES	(1)	(2)	(3)	(4)
	Current ladder		Future ladder
Panel A: MSA-level Analysis
Job creation rate	5.486 (0.978)	5.567 (1.015)	3.588 (0.825)	3.103 (0.682)
Job destruction rate	-3.586 (0.838)	-3.433 (0.870)	-0.158 (0.702)	0.144 (0.668)
Log of income	0.277 (0.077)	0.293 (0.094)	0.221 (0.061)	0.324 (0.073)
Additional MSA controls		x		x
Observations	363	344	363	344
R-squared	0.218	0.246	0.149	0.460
Panel B: Individual-level Analysis
Job creation rate	1.098 (0.395)	1.274 (0.445)	1.068 (0.206)	0.944 (0.220)
Job destruction rate	-0.791 (0.274)	-0.702 (0.306)	0.926 (0.197)	0.987 (0.225)
MSA log of income	-0.197 (0.046)	-0.173 (0.048)	-0.0408 (0.031)	-0.0382 (0.038)
Additional MSA controls		x		x
Individual controls (incl. income)	x	x	x	x
Year and Month F.E.	x	x	x	x
Observations	556,300	461,054	544,228	450,908
R-squared	0.103	0.103	0.094	0.095
Panel C: Panel Analysis
Job creation rate	2.276 (0.316)	1.213 (0.357)	1.690 (0.293)	1.155 (0.349)
Job destruction rate	-0.617 (0.274)	-0.466 (0.314)	-0.647 (0.249)	-0.460 (0.285)
Log of income	0.416 (0.0299)	0.416 (0.0371)	0.266 (0.0304)	0.195 (0.0389)
MSA F.E.		x		x
Additional MSA controls	x	x	x	x
Year and quarter F.E	x	x	x	x
Observations	4,884	4,884	4,884	4,884
R-squared	0.195	0.325	0.145	0.257

The dependent variables are SWB measures from Gallup: columns (1) and (2) use the current Cantril ladder of life whereas columns (3) and (4) use the future ladder of life. Sample years are 2008-2011 and the sample is restricted to working-age respondents.

Columns (1) and (3) regress these life satisfaction measures on the job creation and the job destruction rates, which come, in Panel A and B, from the BDS, and in panel C, from the LEHD which provide data at the quarterly level. All specifications control for income at the MSA-level. Columns (2) and (4) add some MSA-level controls: the unemployment rate, population (in levels), crime rates, and share of African-Americans.

All data sources and variables definitions are in the main text.

Panel A carries a cross-sectional analysis at the MSA-level, where the variables are averaged across years within each MSA. The SWB measures are averaged using the weights attached by Gallup to each respondent.

Panel B carries a repeated cross-section analysis at the individual-level. Month and Year fixed effects are added to the specification as well as individual controls: age, age square, a dummy for being female, 6 dummies for family status, 6 dummies for education, and 4 race dummies (black, asian, white, other or missing), as well as 9 dummies for household income brackets. Regressions are weighted by individual weights attached by Gallup to each respondent.

Panel C shows the results of a quarterly panel analysis at the MSA level. The SWB measures are averaged at the MSA-quarter level using the weights attached by Gallup to each respondent. All regressions include year and quarter fixed effects, quarterly controls for the MSA's average log of income, and the share of African Americans in the MSA, as well as annual controls for population levels and crime rates. Columns (2) and (4) add MSA fixed effects.

Now, consider the magnitudes of the various effects. A one standard deviation increase in the job creation rate is associated with an increase in the current ladder of life of slightly more than half a standard deviation (= 0.018 × 5.567/0.192). A one standard deviation increase in the job destruction rate is associated with a decrease in the current ladder of life of 0.4 (= 0.021 × 3.433/0.192) standard deviations.

Individual-level Results

In Table 3 Panel B we perform individual-level regressions and find qualitatively similar results as in Panel A. Again, we control for many demographic characteristics as well as income brackets. All specifications include year and month fixed effects as well as a control for the MSA's log of income. The job creation and destruction rates vary at the MSA-year level. Standard errors are clustered at the MSA level. Again, columns 2 and 4 are similar to columns 1 and 3 except that additional MSA-level potential confounders are added. We see that these additional controls barely change the coefficient on the job creation and destruction rates. Similar to Prediction 1, the magnitudes are smaller than for the MSA-level results.

Panel Results

In Panel C, we show results of panel specifications using quarterly data on job creation and destruction rates coming from the LEHD, as in Panel C of Table 2. Columns 1 and 3 are without MSA fixed effects, whereas columns 2 and 4 are with MSA fixed effects. All specifications include year and quarter fixed effects as well as MSA-level potential confounders such as share of African Americans, population level, and crime rate. Prediction 2 remains verified in panel analysis, with a positive effect of the job creation rate on SWB and a negative effect of the job destruction rate.

3.5 Testing Prediction 3

In this section, we test Prediction 3: A higher turnover rate increases wellbeing more, whereas a higher job destruction rate decreases wellbeing less, the more generous the unemployment benefits.

The generosity of unemployment insurance (UI) varies at the state level. To avoid the endogeneity of the total number of benefits claimed, we use, as is standard in the literature, the maximum weekly benefit amount as a measure of the state's UI generosity. We normalize it by the average taxable wage. Our results are robust to whether or not we do this normalization. Panel A of Table 4 carries the analysis at the MSA-level whereas Panel B shows the results when using individual level regressions, controlling for the same individual characteristics used in Panels B of Table 2 and 3. The coefficients of interest are that of the interaction term between job turnover and UI generosity, as well as that of the interaction term between job destruction and UI generosity. Indeed we expect the effect of UI generosity to alleviate the negative effect of the job destruction rate by making the risk of unemployment less costly. On the contrary, there is no clear prediction on how UI generosity should interact with the job creation rate. Thus we do not report the interaction terms and main effect of the job creation rate although these variables are included in all the specifications that feature the job destruction rate (i.e., columns 3 and 4). Note that we demean our measure of the state's UI generosity such that the coefficients for the main effect of the job turnover or the job destruction rate show the effect for MSAs located in a state where the UI generosity is at its mean value.

Table 4. Test of prediction 3.

VARIABLES	(1)	(2)	(3)	(4)
	Current ladder
Panel A: MSA-level Analysis
Job turnover rate	0.524 (0.362)	0.662 (0.378)
Job turnover × UI generosity	0.989 (0.422)	0.897 (0.416)
Job destruction rate			-3.661 (0.789)	-3.536 (0.816)
Job destruction × UI generosity			2.357 (1.105)	2.369 (1.113)
UI generosity	-0.288 (0.114)	-0.253 (0.113)	-0.167 (0.128)	-0.137 (0.128)
Additional MSA controls		x		x
Observations	363	344	363	344
R-squared	0.116	0.136	0.237	0.262
Panel B: Individual-level Analysis
Job turnover rate	0.0845 (0.230)	0.209 (0.262)
Job turnover × UI generosity	0.675 (0.310)	0.670 (0.357)
Job destruction rate			-0.794 (0.272)	-0.720 (0.300)
Job destruction × UI generosity			0.620 (0.329)	0.673 (0.372)
UI generosity	-0.198 (0.085)	-0.183 (0.096)	-0.212 (0.083)	-0.200 (0.094)
Individual controls (incl. income)	x	x	x	x
Year and Month F.E.	x	x	x	x
Additional MSA controls		x		x
Observations	556,300	461,054	556,300	461,054
R-squared	0.103	0.103	0.104	0.103

Panel A carries a cross-sectional analysis at the MSA-level, where the variables are averaged across years within each MSA, whereas Panel B carries a repeated cross-section analysis at the individual-level. The first two columns are similar to the columns (1) and (3) of Table 2, and the last two columns are similar to the first two columns of Table 3, except that the creative destruction variables (job turnover, job creation, and destruction rates) are interacted with state-level UI generosity. UI generosity is measured by the average maximum weekly benefit amount over the period 2008-2011 normalized by the average taxable wage in covered employment. The variable is demeaned. We don't report the interaction coefficient for job creation as the interaction of interest is the job destruction one (see main text).

MSA-level Results

We see that the effect of the job turnover rate on SWB at the mean value of UI generosity is positive and that the coefficient of the interaction term between job turnover and the generosity of UI is significantly positive and economically significant (column 1, Table 4 Panel A). A one standard deviation increase in UI generosity increases the positive effect of job turnover by almost 100% (= 0.520 × 0.989/0.524). This remains true in column 2 when we control for the same potential MSA-level confounders as in Table 2 and 3.

This more positive effect of turnover on SWB in MSA located in states with more generous UI is driven by a much less negative effect of the job destruction rate. Indeed we see in column 3 that the direct effect of job destruction on wellbeing is negative but the interaction term between the job destruction rate and the generosity of UI is significantly positive. A one standard deviation increase in our measure of UI generosity reduces the negative effect of the job destruction rate by 33% (= 0.52 × 2.357/3.661). Thus the effect of UI generosity is not only statistically significant but also economically so.

Individual-level Results

The individual-level results of Panel B shows the exact same pattern as the MSA-level results of Panel A. We still have a significantly positive coefficient for the interaction terms of both job turnover and UI generosity (columns 1 and 2) as well as job destruction and UI generosity (columns 3 and 4). Results are barely affected by the addition of potential MSA-level confounders. And the magnitude of the interaction effect is roughly similar to that displayed at the MSA-level: For instance in column 3, a one standard deviation increase in our measure of UI generosity reduces the negative effect of the job destruction rate by 40% (= 0.52 × 0.62/0.794).

3.6 Testing Prediction 4

In the Online Appendix A2.1, we test Prediction 4: A higher turnover rate increases future wellbeing more for more forward-looking individuals.

It is hard to find a direct measure of the discount rate. The literature on this subject found that old individuals, educated individuals, and rich individuals tend to be more patient [Gilman (1976), Black (1983), Lawrance (1991), Warner and Pleeter (2001)]. In the Online Appendix (Table A11) we use age, education, and income to proxy for individuals with different patience levels. This exercise should be seen as a first step, as our data does not provide a measure of patience which is fully convincing on its own. Further tests of this prediction are left to future work to be conducted with better measures of patience.

4 Robustness Checks and Extensions

In this section, we discuss various theoretical and empirical robustness checks and extensions.

4.1 Theory

Transitional Dynamics

In Online Appendix A1.1, we consider the transitional dynamics. More specifically, we look at the dynamic impact of a sudden increase in the entry rate. In particular we show that: (i) following such an increase in the entry rate convergence to the steady-state is fast; (ii) the big change in welfare occurs at the time of the increase in entry rate, and: (iii) the comparative statics on this change are quite similar to the comparative statics on steady-state welfare stated in Proposition 1. Thus there is no loss of insight in restricting attention to the steady state in our analysis in general.

Exogenous Job Destruction

In the current model, the creative destruction rate x generates both job destruction for the workers in incumbent firms and job creation for currently unemployed workers. When the job destruction is fully endogenous, as it currently is, the former effect dominates and the equilibrium unemployment rate u (x) is increases in x as in (6). In Online Appendix A1.2, we extend the model so as to also allow for exogenous job destruction. There we show that the higher the exogenous rate of job destruction, the more will the innovation rate x contribute to reducing unemployment (the latter effect dominates), and therefore the more positive the overall effect of x on equilibrium wellbeing.

Risk Aversion

The analysis in this section can be straightforwardly extended to the case where individuals are risk averse. In Online Appendix A1.3 we show that when agents are risk averse, job loss is perceived more detrimentally than when they are risk neutral. Consequently, there is a range of unemployment benefits for which higher turnover reduces life satisfaction for risk-averse individuals with log preferences whereas it would increase life satisfaction for risk-neutral individuals.

Endogenous Entry

In Online Appendix A1.4, we extend the model to endogenize entry. This in turn enriches our analysis of the relationship between wellbeing and the determinants of creative destruction. In particular, a lower entry cost will have the same effects on wellbeing as the effects of an increase in x, but an increase in the size of innovations will enhance both the growth effect for given x and the creative destruction effect (it will foster x).

Matching Efficiency

In Online Appendix A1.5, we generalize the model by introducing a multiplicative parameter which reflects the efficiency of the matching process. This allows us to look at what happens when the costs of unemployment are incurred over longer expected time periods. We find that as the cost of unemployment has a longer impact (the productivity of matching declines), the negative impact of innovation on wellbeing through unemployment is amplified.

Taxing Labor to Finance UI Benefits

In Online Appendix A1.6, we consider a generalized version of our setting where the unemployment benefit is financed by raising taxes on labor income and corporate profits. We show that our results on unemployment benefit is robust to this generalization.

4.2 Empirics

We perform several robustness checks to confirm the validity of our empirical results on the main predictions. More details are provided in the Online Appendix.

BRFSS Data

First we look at what happens when we use an alternative measure of SWB. Table A2 and A3 use the life satisfaction measure from the BRFSS. In BRFSS, life satisfaction is measured using the question: “In general how satisfied are you with your life?” The possible answers are: “Very satisfied”; “Satisfied”; “Dissatisfied”; “Very dissatisfied”. We recode these answers so that “Very dissatisfied” corresponds to grade 1 and “Very satisfied” to grade 4.

Table A2 tests for Prediction 1 and shows that, even when using the BRFSS measure of SWB, the effect of MSA job turnover on wellbeing is more positive when we control for the unemployment rate than when we do not. The difference is statistically significant at the 1% level. Table A3 tests for Prediction 2 and confirms a significant positive effect of job creation on life satisfaction and negative effect of job destruction, also with this alternative measure of our left-hand side variable.

Effect of the 2008 Crisis

In the BRFSS, the life satisfaction measure starts in 2005 instead of 2008 for the Gallup variables. Thus BRFSS allows us to check that our results are robust to restricting attention to different subperiods. In particular, to deal with the concern that the period post 2008 is a period following a major recession, where all kinds of other things were going on in developed economies, in Table A4 and A5 we decompose the overall BRFSS sample period into the subperiods 2005-2007 and 2008-2010. Table A4 tests for Prediction 1 and Table A5 tests for Prediction 2.

In Table A4 we find that the effect of turnover on life satisfaction is less positive during the crisis years than in the period before, but this difference disappears when controlling for unemployment. In Table A5 we find that job creation (resp. job destruction) has a more positive (resp. negative) effect on wellbeing before the crises years.

Alternative Definition of Creative Destruction

We then look at what happens when we use an alternative measure of creative destruction. In Tables A6 and A7 we use the measure that comes from the LEHD, which we also use in our main panel analysis because of its quarterly nature. Table A6 reproduces the baseline MSA cross-sectional specification of Panel A Table 2 (testing for Prediction 1) but using this LEHD measure of job turnover, whereas Table A7 reproduces the baseline MSA cross-sectional specification of Panel A Table 3 (testing for Prediction 2) but using the LEHD measure of job creation and job destruction. We see that both Predictions 1 and 2 are still verified in cross-section when we use these other measures of creative destruction. Indeed the effect of job turnover as measured by the LEHD on the current ladder of life is more positive when we control for the unemployment rate than when we do not and the difference is statistically significant at the 5% level. The effect on wellbeing of the job creation rate as measured by LEHD is significantly positive and that of the destruction rate is significantly negative.

Non-linearity of Unemployment

In Table A8, we show that Prediction 1 is robust to controlling non-linearly for the unemployment rate. Indeed we introduce a cubic polynomial of the unemployment rate instead of just the unemployment rate to the baseline specification of Panel A Table 2. We still have that the effect of job turnover in column 2, i.e., when controlling for the unemployment rate, here non-linearly, is statistically different at the 1% level from the effect of job turnover in column 1, when we do not control for unemployment. Since the unemployment rate does not play any role in Prediction 2, we only test the robustness to non-linear control for the unemployment rate for Prediction 1.

Bartik Analysis

The last and important robustness check we perform, in Tables A9 and A10, aims at alleviating a potential endogeneity concern. Indeed, to abstract from the effects of local changes in industry composition, or from the effects of purely local shocks that could get mixed up with variations in local turnover, we construct a “predicted measure” (or Bartik-type measure) of creative destruction as follows:

$${\widehat{CD}}_{m,t}=\sum _j{\omega }_{j,m,2004}\times C{D}_{j,m,\mathit{\text{USA}},t}$$

For each MSA m in quarter-year t : (i) the predicted level of Creative Destruction, ${\widehat{CD}}_{m,t}$, is computed by taking a weighted average of countrywide sectoral turnover measures in quarter-year t; (ii) CD_j,m,USA,t is the country-wide average creative destruction in sector j leaving out MSA m; (iii) the weights ω_j,m,2004 are determined by the sectoral structure in the MSA in 2004 (sectors are 2 digit NAICS).

Thus, we reproduce the MSA-level quarterly panel regressions of panels C in Tables 2 and 3 but replacing the direct local turnover variable by its predicted value ${\widehat{CD}}_{m,t}$. Standard errors are still clustered at the MSA level.²¹ The results turn out to be quite similar when using the predicted measure of turnover instead of the actual quarterly turnover rate as the right-hand side variable. In particular, for Prediction 1, the coefficient for job turnover is larger when we control for unemployment than when we do not and the difference is significant at the 1% level. The coefficient remains positive and significant when we add MSA fixed-effects. Interestingly the MSA fixed-effects do not make the coefficient decrease as much as when using the actual turnover rate as the main right-hand side variable. Table A10 shows that the effect of job creation and job destruction, when captured by these predicted measures, are still significantly positive and negative, respectively.

5 Conclusion

In this paper we have analyzed the relationship between turnover-driven growth and SWB, using cross-sectional MSA level US data. We have first built a Schumpeterian model of growth and unemployment to make predictions on how job and firm turnover affect wellbeing under various circumstances. Our main empirical findings are consistent with the theory: namely: (i) the effect of creative destruction on wellbeing is unambiguously positive if we control for unemployment, less so if we do not; (ii) job creation has a positive and job destruction has a negative impact on wellbeing; (iii) job destruction has a less negative impact in MSA within states with more generous unemployment insurance policies; (iv) job creation has a more positive effect on individuals that are more forward-looking. We see these findings not just as a test of the Schumpeterian theory of growth, creative destruction, and unemployment, but also of the usefulness of current and future wellbeing measures.

This is the first step of a broader research project on innovation-led growth and wellbeing. A first avenue forward could be to use a similar combination of the theory and of cross-section analysis to investigate other potential determinants of wellbeing and compare them with the determinants of (per capita) GDP growth. A second extension would be to look at how the relationship between turnover and wellbeing is affected by individual characteristics and by characteristics of labor markets and labor market policy (e.g., training systems and availability of vocational education). A third extension would be to look for policy shocks (e.g., labor market reforms) that may affect the relationship between creative destruction and wellbeing. These and other extensions of the analysis in this paper are left for future research.

A Proof of Lemma 1

The output in this economy is

$$ln{Y}_t={\int }_{j\in \mathcal{J}}ln{A}_{jt}dj\equiv (1-u)ln{\overline{A}}_t$$

Then after a small time interval Δt:

$$\begin{array}{c}ln{Y}_{t+\mathrm{\Delta }t}={\int }_{\mathcal{J}}[x\mathrm{\Delta }t\times 0+(1-x\mathrm{\Delta }t)ln{A}_{jt}]dj+{\int }_{{\mathcal{J}}^{\prime }}[\frac{m}{v}\mathrm{\Delta }tln(1+\lambda ){\overline{A}}_t+(1-\frac{m}{v}\mathrm{\Delta }t)\times 0]dj\\ =(1-x\mathrm{\Delta }t)(1-u)ln{\overline{A}}_t+u\frac{m}{v}\mathrm{\Delta }tln(1+\lambda ){\overline{A}}_t\\ =[1-u]ln{\overline{A}}_t+m\mathrm{\Delta }tln(1+\lambda )\end{array}$$

Hence we can find the growth rate as

$$g=\underset{\mathrm{\Delta }t\to 0}{lim}\frac{ln{Y}_{t+\mathrm{\Delta }t}-ln{Y}_t}{\mathrm{\Delta }t}=mln(1+\lambda )$$

B Derivation of Equation (10)

Note that using the fact that in steady state Ė = gE and U̇ = gU, and after subtracting the second equation from the first:

$$(r-g)(E-U)=BY+(1+x)(U-E),$$

where B = βπ − b.

This yields:

$$E-U=\frac{BY}{r-g+1+x}.$$

Then, substituting for (E − U) in the above asset equations (8) and (9), yields:

$$U=[bY+\frac{BY}{r-g+1+x}]\frac{1}{r-g};\text{and}E=[\beta \pi Y-\frac{xBY}{r-g+1+x}]\frac{1}{r-g}.$$

so that, after substituting for E and U in the expression for W, and using the fact that in equilibrium u = x/(1 + x), we get:

$$W=\frac{W}{r-g}[\beta \pi -\frac{xB}{1+x}].$$

C Derivation of Equation (12)

Recall that:

$$W=uU+(1-u)E,$$

where E and U are expressed in (8) and (9). Now, using the fact that m(u, v)/u = (1−u)x/u and that in steady state Ė = gE and U̇ = gU, we obtain:

$$E-U=\frac{BY}{r-g+x/u}.$$

Substituting for (E − U) in the asset equations (8) and (9), yields:

$$U=[bY+\frac{[(1-u)x/u]BY}{r-g+x/u}]\frac{1}{r-g},$$

and

$$E=[\beta \pi Y-\frac{xBY}{r-g+x/u}]\frac{1}{r-g}.$$

so that:

$$W=\frac{Y}{r-g}[ub+(1+u)\beta \pi ].$$