Household Production and the Elasticity of Marginal ...vti.diva-portal.org/smash/get/diva2:952369/FULLTEXT01.pdf · Household Production and the Elasticity of Marginal Utility of

HOUSEHOLD PRODUCTION AND THE ELASTICITY OF MARGINAL UTILITY OF CONSUMPTION

Disa Thureson

The Swedish National Road and Transport Research Institute (VTI), Division of Transport Economics,

and Örebro University School of Business, Sweden

CTS Working paper 2016:10

Abstract Chetty (2006) developed a new method of estimating the Elasticity of Marginal Utility of Consumption (EMUC) from observed work time responses to wage changes and derived an upper bound of 2 for this parameter. Here I show that the omission of household production in Chetty’s model may lead to bias, and perform a numerical sensitivity analysis of Chetty’s results in this respect. I develop a new model that includes household production from which I derive new, unbiased EMUC formulas. I offer empirical estimates based on current evidence of the included parameters, suggesting a lower bound for EMUC of about 0.9.

Keywords: Elasticity of marginal utility of consumption; Household production; Labor supply; Coefficient of relative risk aversion; Consumer behavior

JEL codes: D10, D60, J22

Centre for Transport Studies SE-100 44 Stockholm Sweden

www.cts.kth.se

http://www.cts.kth.se/

Household Production and the Elasticity of Marginal Utility of Consumption

2

1 INTRODUCTION The link between individuals’ consumption and wellbeing is at the core of economic analysis, from both a positive and a normative perspective. One way to estimate this relationship is to study the tradeoffs people make between leisure and consumption. Here I suggest an approach that adds household production in this analysis. The Elasticity of Marginal Utility of Consumption (EMUC) is a means for measuring how utility changes with consumption and is of great importance for understanding consumer behavior as well as for policy evaluation. In for instance evaluation of climate change policies, EMUC is one of the most influential parameters in the calculation of the social cost of carbon (Anthoff et al., 2009). Using the Ramsey equation, discounting the costs of future disasters should accounts for both the pure rate of time preference and the growth rate of consumption over time; the two components are added after weighting future consumption with EMUC (see, e.g., Baum 2009). This weight can also be interpreted as the relative significance of consumption of current vs. future generations, i.e., a distributional interpretation of the concept. In analyses of issues related to risk, the EMUC coefficient corresponds to a commonly used interpretation of the coefficient of relative risk aversion. In a review of empirical studies of EMUC estimates, Evans (2005) reports estimates that are mostly in the range of 1–2. While previous analyses used the pattern of savings or information about, e.g., the demand for food for approximating EMUC, Ray Chetty (2006) demonstrated that observations of how working time responds to wage changes could be an alternative approach. He argued that if people increase their working time as their wage increases, this implies that on average EMUC is less than unity. Through the Slutsky decomposition of labor supply, there are two effects of wage changes on the demand for leisure: one income effect, where a higher wage increases income and leads to higher demand for leisure, and one substitution effect, which increases the relative price of leisure at the margin and hence reduces the leisure demand. If utility from leisure is also independent of consumption level and if market work constitutes the only source of income, an EMUC value equal to unity means that these two effects cancel each other out. If instead EMUC is less than unity, the substitution effect dominates (leisure demand decreases), but if EMUC is greater than unity, the income effect dominates (leisure demand increases). Assuming that reductions (increases) in leisure demand are realized through working time increases (reductions), EMUC can be estimated from evidence on working time responses to wage changes. The starting point of Chetty’s approach is that the (uncompensated) labor supply elasticity with respect to wage is positive on average and hence the average EMUC is lower than unity. If, in addition to wage income, there is unearned income and if consumption and labor are complementary, this upper bound is relaxed. He derived EMUC formulas under such circumstances and


3

computed corresponding numerical values based on labor supply elasticity estimates, an concluded that the existing empirical evidence implies that EMUC must be less than 2: “The intuition for this tight bound is simple: if the marginal utility of wealth diminishes rapidly, why don’t people choose to work much less when their wages rise?” (Chetty 2006, p. 1830). Here I propose an approach to estimate EMUC that is based on an additional basic feature of people’s day-to-day tradeoffs made as a result of wage level changes. The idea is that individuals may respond to wage increases not only via adjustment of their labor supply but also by reducing the time spent on housework or other forms of unpaid work. For example, high-wage individuals can reduce time spent in household production by buying services from low-wage individuals. I argue that effective leisure time, rather than time not spent on paid work in the formal sector, is what is relevant when making inferences about the utility function. The main aim of the present study is therefore to investigate how considerations of household production influence theoretical and empirical conclusions regarding feasible long-term, average values of EMUC. The analysis is restricted to the use of intensive margin responses of labor supply, and I will not explicitly model intertemporal substitution of individuals1. To this end, I develop a model with household production that is used to derive some properties of the bias invoked by the omission of household production. I further derive EMUC equations for two special cases based on the wage elasticity of labor supply. One of these equations can also be used to estimate a lower bound for EMUC under more general conditions. I use the uncompensated wage elasticity of labor supply rather than – as in Chetty’s final expression – the compensated wage elasticity in combination with the income elasticity of labor supply. EMUC estimates from this new specification are less sensitive to the magnitude of the assumed proportion of unearned income to earned income than in the previous formulation. In addition, this specification allows utilization of labor supply elasticity estimates based on other empirical sources, notably time series and cross-country data on mean working times and mean wages, for which it is possible to control for household production to some extent. The main conclusion of the analysis is that when household production and market work can be substituted, omission of household production in an EMUC equation will generally result in biased estimates. Examples of this are shown in a sensitivity analysis of the differences in results with and without inclusion of household production. In addition, the sensitivity analysis shows that the results in Chetty (2006) are highly sensitive to the assumed proportion of unearned income to earned income. The main results of the present study suggest a lower bound of the average EMUC close to unity (about 0.9 at the

1 However, for the numerical estimation I will consider two cases where consumers either maximize lifetime utility or within-period utility.


4

lowest). In addition, I compute point estimates in the range of 0.9–1.6, but unfortunately these latter results should be considered more as an illustration of the proposed method than a final conclusion, since some of the underlying assumptions are based on estimates from a single old, and possibly biased, data source. The next section starts with a description of the original model and continues with my extension of the model. Section 3 reviews the literature on the relevant parameters. In Section 4, results from the numerical analysis are presented followed by a discussion in Section 5 and concluding remarks in Section 6.

2 THEORY

2.1 Chetty’s model

In Chetty (2006), an agent faces the following utility maximization problem during 𝑇 time periods: max

𝑐𝑡,𝑀𝑡

𝑈(𝑐1, … , 𝑐𝑇 , 𝑀1, … , 𝑀𝑇)

(1) s.t. 𝑝1𝑐1 + ⋯ + 𝑝𝑇𝑐𝑇 = 𝑦 + 𝑤(𝜃1𝑀1 + ⋯ + 𝜃𝑇𝑀𝑇) where 𝑐𝑡 is consumption in period 𝑡, 𝑝𝑡 is price level of consumption in each time period, 𝑦 is unearned income (endowment), 𝑤 is mean wage, and 𝜃𝑡 is wage share in each time period. 𝑀𝑡 is the time spent in market work (labor supply) out of the time at one’s disposal in each time period 𝑡. This means that market work is the opposite of leisure time and that utility is decreasing in market work when consumption is held constant. Chetty used a two-stage formulation to reduce expression (1) to: max𝑐,𝑀 𝑢(𝑐, 𝑀) s.t. 𝑐 = 𝑦 + 𝑤𝑀, (2)

where 𝑢(𝑐, 𝑀) = max

𝑐𝑡,𝑀𝑡

𝑈(𝑐1, … , 𝑐𝑇 , 𝑀1, … , 𝑀𝑇),

s.t. 𝑝1𝑐1 + ⋯ + 𝑝𝑇𝑐𝑇 = 𝑐 and 𝜃1𝑀1 + ⋯ + 𝜃𝑇𝑀𝑇 = 𝑀. He further showed that when 𝑦 = 0 and consumption and labor are additively separable in the utility function, the following result holds:

𝜕𝑀

𝜕𝑤> 0

⇔ 𝜂 < 1. (3)

where 𝜂 is EMUC. The intuition is that an increase in wage implies both an increase in income and an increase in the price of leisure relative to the consumption numeraire. When EMUC is 1 (and market work constitutes the only source of income and the utility of leisure is independent of consumption), the two effects cancel out and leisure demand remains constant. When EMUC is less than 1, the price effect dominates and leisure demand decreases.


5

Chetty then broadened the model to allow for complementarity between market work and consumption and for unearned income to make it possible to empirically derive an upper bound of EMUC. He defined the level of complementarity between market work and consumption as:

lim∆𝑀→0

∆𝑐

𝑐/

∆𝑀

𝑀. (4)

This can be interpreted as the compensated elasticity of consumption with respect to labor supply. This would correspond to the consumption response due to a part-time unemployment shock that is fully compensated, for instance, by unemployment insurance.2 For simplicity, I will denote the compensated elasticity of consumption with respect to working time as 𝜀��𝑀. The resulting expression of EMUC was equation (9) in Chetty (2006):

�� =(1+

𝑤𝑀

𝑦)

−𝜀𝑀𝑦

��𝑀𝑤

(1−(1+𝑦

𝑤𝑀)��𝑐𝑀)

, (5)

where 𝜀��𝑤 is the compensated elasticity of market work with respect to wage, defined by:

𝜀��𝑤 =𝜕𝑀

𝜕𝑤

𝑐∙

𝑤

𝑀, (6)

and:

𝜕𝑀

𝜕𝑤

𝑐=

𝜕𝑀

𝜕𝑤− 𝑀

𝜕𝑀

𝜕𝑦, (7)

according to the Slutsky decomposition of compensated labor supply. Equation (5) consist of the following three factors: an unearned income factor,

(1 +𝑤𝑀

𝑦), an income response factor,

−𝜀𝑀𝑦

��𝑀𝑤, and a complementarity factor,

1

(1−(1+𝑦


. Intuitively, the core part of equation (5) is the income response

factor, which is balanced by the proportions between earned and unearned income through the unearned income factor. Lastly, the possibility of complementarity (or substitutability) between market work and consumption is controlled for through the complementarity factor (when 𝜀��𝑀 = 0 the complementarity factor collapses to 1). A few notes can be made regarding the income response factor. First, the elasticity of market work w.r.t. unearned income, i.e. 𝜀𝑀𝑦, is generally assumed to be negative, so the numerator is

positive. Also, compensated elasticity of labor supply w.r.t. wage, i.e. 𝜀��𝑤, is assumed to be positive, and hence the income response factor is positive. One can also see that EMUC estimates (through the income response factor) are proportional in size to 𝜀𝑀𝑦 and inversely proportional to 𝜀��𝑤.

2 If a person wants to consume more in periods when she works more relative to periods when she has more leisure, there exists a positive complementarity between labor and consumption according to this definition.


6

I will now review Chetty’s results from both a theoretical and empirical point of

view. The unearned income factor (1 +𝑤𝑀

𝑦) is obviously sensitive to small

values of 𝑦 in relation to 𝑤𝑀 (as 𝑦 approaches zero, the unearned income factor goes to infinity). As I will show in Section 3 using disaggregate data, (proportionally) small values of 𝑦 seem more plausible than the values suggested in Chetty’s study. I will further show that since potential substitution between market work and household production is not controlled for in Chetty’s model, both the income response factor and the complementarity factor may be biased. The bias that arises from the omission of household production will typically lead to an underestimation of the income response factor, whereas for the complementarity factor both underestimation and overestimation can occur. Related to the issue of omission of household production, I will also consider the empirical possibility that the uncompensated wage elasticity of labor supply is

negative instead of positive, i.e. that 𝜕𝑀

𝜕𝑤< 0. Through the inequality in equation

(3), this will have implications for the possibility to estimate an upper or a lower bound.

2.2 Introducing household production

I now define a model that is similar to expression (2) but includes household production. The agent faces the following maximization problem: max

𝐿,𝑀,𝐻𝑢(𝑐, 𝐿), (8)

s.t. = 𝑦 + 𝑤𝑀 + 𝑔(𝐻), 𝜏 = 𝐿 + 𝑀 + 𝐻, where 𝑔′(𝐻) > 0, 𝑔′′(𝐻) < 0, where 𝑐 is consumption, 𝐿 is leisure time, 𝑦 is unearned income (endowment), 𝑤 is wage, 𝜏 is total time at disposal, 𝑀 is time spent on market work (labor supply), and 𝐻 is time spent on household production. 𝑔(𝐻) is the household production function and represents the consumption value of total household production for the agent. It captures substitution between market work and household production through the following three channels: substitution between market goods and home produced goods, labor division in household production (hiring of staff or work sharing with other household members), and investments in household production capital. I restrict the model by defining 𝜂 for the normalization3 𝑔(𝐻0) = 0.

3 The reason for the normalization is twofold. Most importantly, in the standard definitions of 𝜂, only market goods and services are included in the consumption numeraire. From this perspective, the total value of the household production needs to be normalized to zero if EMUC values are to be


7

Leisure time includes all activities other than market work and household production. The focus is on intensive margin responses of labor supply and therefore I restrict the model to agents that to some extent participate in market work. It is also assumed that some housework tasks are not feasible to substitute for market work, that is, lim

𝐻→0 (𝑔′(𝐻)) = ∞.4 It follows that an optimal

solution must satisfy: 𝑐 > 0, (9a) 𝐿 > 0, (9b) 𝑀 > 0, (9c) 𝐻 > 0. (9d) Figure 1 gives a visual interpretation of the optimization problem defined in equation (8).

Figure 1: Optimization problem of the agent.

The agent has 𝜏 time to allocate between leisure time, 𝐿, market work, 𝑀, and household production, 𝐻. She first optimizes the proportion of household production in relation to market work, which will be the point where the slopes of 𝑔(𝐻) and 𝑤𝑀 equate (equilibrium 1). She then chooses the optimal level of

comparable in size to the standard definition. The normalization also implies simplification of the model. 4 Another possible formulation of a model that includes household production is max

𝐿,𝑀,𝐻𝑢(𝑐, 𝐿, 𝑔),

where 𝑐 = 𝑦 + 𝑤𝑀 and 𝑔 = 𝑔(𝐻), which is only partly tradable in the market. However, as long as g and c are additive, this does not make a considerable difference, since 𝑔(𝐻) is general enough to allow for extremely low substitutability between g and c for some tasks, which in practice will mean that these tasks will not be possible to substitute for market work.

𝜏

𝑐

𝑔(𝐻)

𝐻 𝐿 𝑀

Indifference

curve

𝑐 = 𝑔(𝐻0) = 0

𝑐 = 𝑤𝑀

1

2

𝑤𝑀

𝐻0 𝐿0


8

consumption in relation to leisure, which is at the point where the slope of the indifference curve equates the slope of 𝑤𝑀 (equilibrium 2). I now put forward some propositions regarding solutions for the utility-maximizing behavior of the agent described by expression (8) and expression (9) and regarding the estimator implications of that behavior. Proposition 1 describes how the chosen level of household production is influenced by wage changes. This is important for the understanding of the mechanisms in the model and for the interpretation of the results later on. For a derivation of Proposition 1, see Appendix 1. Proposition 1: The derivative of time spent in household production with respect to wage

is inversely proportional to the second derivative of the household production function, and hence negative:

𝜕𝐻

𝜕𝑤=

1

𝑔′′(𝐻)< 0. (11)

There are two limiting cases of equation (11) w.r.t. 𝑔′′(𝐻):

lim𝑔′′(𝐻)→0

(𝜕𝐻

𝜕𝑤) =

−∞, (12a)

lim𝑔′′(𝐻)→−∞

(𝜕𝐻

𝜕𝑤) = 0. (12b)

Equation (12a) represents a case of perfect substitution between household production and market production. In this case there is no internal solution, but instead either of the two corner solutions of 100% household production or 100% market work will arise, which violates assumption (9c) and assumption (9d), respectively. The implication for an internal solution is that if the level of concavity of the household production function is low, a small increase in wage can cause a great shift between unpaid and paid work. Equation (12b) implies that when the level of concavity of the household production function is high, the shift between household production and market production will be small and can be approximated to zero. Proposition 2 provides an estimator of EMUC in a model that includes household production. For the derivation of Proposition 2, see Appendix 2. Proposition 2: In a model with household production, EMUC can be expressed as:

𝜂 =

(1+𝑤𝑀

𝑦)

−𝜀𝑀𝑦𝑃𝑀

��𝑃𝑤

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

(13)

where:


9

𝑃 = 𝑀 + 𝐻 is total production time,

𝜀��𝑤 =𝜕𝑃𝑐

𝜕𝑤∙

𝑤

𝑃 is the compensated production time elasticity w.r.t. wage,

where 𝜕𝑃𝑐

𝜕𝑤=

𝜕𝑀𝑐

𝜕𝑤+

𝜕𝐻

𝜕𝑤 ,

and 𝜀��𝐿 =𝜕𝑐𝑐

𝜕𝐿∙

𝐿

𝑐 is the compensated consumption elasticity w.r.t. leisure.

The difference between equation (13) and equation (5) is that, loosely speaking, household work is now also included in the elasticities that involve market work in Chetty’s model. More precisely, when making inference about the utility function (with respect to tradeoffs between consumption and leisure), the relevant variable is how supply in total production time is changed as wage

changes, i.e. the relevant variable is 𝜕𝑃

𝜕𝑤=

𝜕𝑀

𝜕𝑤+

𝜕𝐻

𝜕𝑤. When

𝜕𝑀

𝜕𝑤 is used as a proxy

for 𝜕𝑃

𝜕𝑤, total production time responses will be positively biased by the size of

𝜕𝐻

𝜕𝑤.

Proposition 3 describes properties of the bias caused by omission of household production in Chetty’s model. For the derivation of Proposition 3a and Proposition 3c, see Appendix 3. Proposition 3a: The error factor of an EMUC estimator given by Chetty’s model, without

household production, and defined as the ratio between the estimator given by Chetty’s model and the true value, is:

��

𝜂= (1 +

𝜕𝐻

𝜕𝑤𝜕𝑀𝑐

𝜕𝑤

) ∙

1−(1+𝑦

𝑤𝑀)��𝑐𝑀∙

𝜕𝑀

𝜕𝑤/(−

𝜕𝐿

𝜕𝑤)

1−(1+𝑦

𝑤𝑀)��𝑐𝑀

. (14)

Proposition 3b follows directly from equation (14): Proposition 3b: If there is approximately no change in time spent on household production

in response to a wage change, Chetty’s model without household production will yield an unbiased estimator of EMUC:

𝜕𝐻

𝜕𝑤≈ 0

⇒

𝜂

𝜂

≈ 1 (15)

In order to facilitate sensitivity analysis, it is convenient to split equation (14) into two factors:

𝜂

𝜂

= (𝜂

𝜂

)𝐴

∙ (𝜂

𝜂

)𝐵

, (16a)

where

(𝜂

𝜂

)𝐴

= 1 +𝜕𝐻


𝜕𝑤

, (16b)


10

is the income response error factor, and

(𝜂

𝜂

)𝐵

=1−(1+

𝑦


𝜕𝑀

𝜕𝑤/(−

𝜕𝐿

𝜕𝑤)

1−(1+𝑦


, (16c)

is the complementarity error factor. Proposition 3c: In general, the income response error factor will be within the range zero

to unity:

(𝜂

𝜂

)𝐴

= 1 +𝜕𝐻


𝜕𝑤

∈ (0,1), (17a)

with

lim|𝜕𝐻

𝜕𝑤|→

𝜕𝑀𝑐

𝜕𝑤

(𝜂

𝜂

)𝐴

= 0. (17b)

Proposition 3c implies that the size of household production responses in relation to the compensated labor supply responses w.r.t. wage determines the

size of the bias given by (𝜂

𝜂

)𝐴

. The interpretation is that the bias is driven by the

share of the compensated labor supply response that is caused by a counteracting change in household production, as a change in labor supply cannot be identified as a substitution between leisure and consumption. Proposition 3d follows directly from equation 16c: Proposition 3d: The size of the percental error originating from the complementarity error

factor, i.e.:

|(𝜂

𝜂

)𝐵

− 1| =

||

(1+𝑦

𝑤𝑀)��𝑐𝑀(1−

𝜕𝑀𝜕𝑤

(−𝜕𝐿𝜕𝑤

))

1−(1+𝑦

𝑤𝑀) |

|, (18)

is an increasing function5 of |𝜕𝐻

𝜕𝑤|, given a fixed value of

𝜕𝑀

𝜕𝑤.

Note that in the special case when utilities from leisure and consumption are

additive, i.e., 𝜕2𝑈

𝜕𝑐𝜕𝐿= 0, it follows that 𝜀��𝑀 = 0 and (

𝜂

𝜂

)𝐵

= 1, and the relative bias

5 Remember that

𝜕𝐿

𝜕𝑤= − (

𝜕𝐻

𝜕𝑤+

𝜕𝑀

𝜕𝑤).


11

of �� will reduce to the more convenient expression 𝜂

𝜂

= (𝜂

𝜂

)𝐴

= 1 +𝜕𝐻


𝜕𝑤

≤ 1, that

is EMUC will be underestimated when household production is omitted.

Since the complementarity error factor, i.e. (𝜂

𝜂

)𝐵

generally is a rather complex

expression, its implications will be investigated in more detail numerically6 in Section 4.1. Propositions 3b, 3c, and 3d show that there will be an estimator bias if household production changes as wage changes. According to Proposition 1, changes in household production w.r.t. wage are in turn determined by the concavity of the household production function. As can be seen from combining Propositions 1, 3b, 3c, and 3d, if substitutability between household production and market work is low then the bias will be small and vice versa. Above, I have now put forward some general propositions on the bias that arises in a model that does not include household production, if, in fact, household production is important. In the subsequent Section 2.4, I will derive formulas of more practical use, which I will later (in Section 4) estimate using empirical evidence (from Section 3).

2.3 New EMUC formulas

I will now propose new point estimators for two special cases. One of these formulas can also be used to estimate a lower bound of EMUC under some circumstances.

Proposition 4 presents a formula of EMUC for the special case 𝜕𝑃

𝜕𝑤= 0, which

coincides with the limit of a lower bound when either 𝜕𝑃

𝜕𝑤< 0 or the more

restrictive assumption 𝜕𝑀

𝜕𝑤< 0 holds. For the derivation of Proposition 4, see

Appendix 4. Proposition 4a: If there is no change in total work supply in response to a wage change, i.e.

𝜕𝑃

𝜕𝑤= 0, then EMUC can be estimated as:

𝜂 =(1+

𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

. (19)

Proposition 4b:

6 Because the formula of (

𝜂

𝜂

)𝐴

is linear in 𝜕𝐻

𝜕𝑤/

𝜕𝑀𝑐

𝜕𝑤, an analogous analysis for (

𝜂

𝜂

)𝐴

would be trivial and

will therefore not be performed.


12

If instead the response is a decrease in total work supply, i.e. 𝜕𝑃

𝜕𝑤< 0, or

more restrictively, in market work, i.e. 𝜕𝑀

𝜕𝑤< 0, then equation (19) can be

used to estimate the limit of a lower bound of EMUC. Proposition 5 presents a formula of EMUC for the special case of additive

utilities from leisure and consumption, i.e. when 𝜕2𝑈

𝜕𝑐𝜕𝐿= 0. For the derivation of

Proposition 5, see Appendix 5.

Proposition 5:

If leisure and consumption are additively separable in the utility function,

i.e. 𝜕2𝑈

𝜕𝑐𝜕𝐿= 0, then EMUC can be estimated as:

𝜂 ≈1+

𝑦

𝑤𝑀

1+𝜀𝑀𝑤(1+

𝜕𝐻𝜕𝑤𝜕𝑀𝜕𝑤

)

. (20)

3 EMPIRICAL EVIDENCE FOR PARAMETER VALUES This section reviews the empirical evidence for the uncompensated labor supply elasticity; for changes in shares of market work, household production, and leisure; for the compensated consumption elasticity w.r.t. leisure; for the proportion of market work to leisure; and for the proportion unearned income, as these are the parameters included in the formulas in Section 2.4.

3.1 Uncompensated labor supply elasticity

Evers et al. (2008) summarized 209 estimates of the uncompensated labor supply elasticity w.r.t. wage that were obtained from 30 different studies based on individual data from various OECD countries. A vast majority of the estimates were positive, but the ranges were rather wide, from −0.08 to 2.87 for women and from −0.24 to 0.45 for men, although the authors only selected studies with high credibility8. Moreover, as Blundell and MaCurdy (1999) and Evers et al. (2008) demonstrated, the estimation of such elasticities is a demanding task involving numerous potential pitfalls. For example, Evers et al (2008) explained, “The standard model of labour supply does not distinguish between the effect of wages and taxes on the decision to participate (the extensive margin) and the

7 Values above 2 come from studies performed during the period 1966–1979. As women’s participation rate has changed significantly since then, these values might be outdated. After 1980, the maximum value is 1.2. 8 The authors have selected studies from reputable journals and studies that do not use the OLS method because this method is known to produce unreliable results.


13

decision regarding hours worked (intensive margin)” (p. 29). They further cited Mroz (1987) who found that “when this effect is neglected, the estimated wage elasticity is biased upwards, because hours of work conditional on participation are relatively inelastic with respect to the net wage, while the participation decision is quite elastic with respect to the net wage.” (p. 29). Evers et al. (2008) showed that the estimation results from various studies are sensitive to the specification of participation for men and are sensitive to the specification of family size and children for women. On the macro-economic level, there are empirical indications that when wages in society increase, working time decreases. High-income countries tend to have shorter working times (excluding the unemployed) than low-income countries (see for example, Figure 1 in Morley et al. (2010)). Burgoon and Baxandall (2004) and Huberman and Minns (2007) presented regression results of working time using mean income9 as an independent variable for 14–18 OECD countries using time series data for 22 and 31 years, respectively10 (with about a century in differences between the two studies).11 Both studies showed a significant negative relationship. Huberman and Minns (2007) provided estimates of uncompensated labor supply elasticity in the range −0.11 to −0.16 for the full sample. However, none of these studies controlled for the diminishing productivity of workers with long working hours. Pencavel (2014)12 suggested that marginal productivity is rather constant up to about 50 hours/week for factory workers. According to Huberman and Minns (2007), that threshold was passed (from above) around 1920 for all the OECD countries considered, and this means that later studies of the uncompensated wage elasticity are more reliable in this respect. Vandenbroucke (2009) estimated a simple general equilibrium model calibrated on base-level data describing the US economy in the year 1900 to make an ex-post prediction of the development of labor supply during the first half of the 20th century. The model predicted 82% of the observed decline in working time between 1900 and 1950, and it was concluded that most of the decline was caused by the increases in wages.

9 Burgoon & Baxandall (2004) used GDP/capita and Huberman & Minns (2007) used wage. 10 The results of Huberman & Minns (2007) were estimated using data from 1870–1900, and Burgoon & Baxandall (2004) used data from 1979–2000. 11 Note that in Burgoon & Baxandall (2004), estimation of the relationship between working times and income was not the primary aim of the study, but GDP/capita was used as a control variable. Also note that Huberman and Minns (2007) studied firm-level data with multiple controls associated with the country in question, and hence it was not a strict macroeconomic study, but rather a semi-macroeconomic study. In addition, of the around 20,000 observations included, the vast majority were from the US. They controlled for sex of workers in each firm-level observation. Examples of country/time-specific controls included, for example, the average age of the population, the capital to labor ratio, and the state of the business cycle. Burgoon and Baxandall (2004) controlled for unemployment, tax rate, inequality, and political regime. 12 Note that the results from Pencavel (2014) were not peer-reviewed and the sample was rather small (<200 observations).


14

3.2 Changes in household production and leisure

The theoretical analysis in Section 2.2 indicates that it may be important to consider household production if EMUC is to be estimated from labor supply response data. Although the use of macro data may reduce bias linked to the omission of household production, it does not eliminate it. Even though work sharing may be controlled for in the macro case, the problem of substituting household production time for household production capital remains.13 The studies concerning the development of household production and leisure are based on data from various survey studies that is linked together. A complicating factor is that sometimes the division between household production and leisure is not straightforward, for example, when it comes to child care and education. Also, there have been structural changes such as changes in demographics (an older population with fewer children), dramatically increased gender equality (both in household production and in market work, according to Ramey and Francis (2009)), and longer average educations. Two recent studies of the development of ∆𝐻 over time (as wages have increased) in the United States came to partly different conclusions, although using similar methods14. Aguiar and Hurst (2007), whose analysis was restricted to respondents aged 21–64 years who were neither students nor retirees, found substantial decreases in both market work and non-market work and increases in leisure between 1965 and 2003.15 Ramey and Francis (2009), who estimated time spending for all demographic ages for the years 1900–2005, found the same trends as Aguiar and Hurst for the total population. But contrary to common expectations16, Ramey and Francis (2009) found that the allocation of time spent on market work, household production, and leisure averaged over both sexes in the age group 25–54 years was basically the same in 2005 as a century ago17. Table 1 shows the discrepancy18 in leisure development between the two studies during the period 1965–2003.

13 An additional issue is that GDP/capita is negatively correlated with the size of the informal sector (Ihrig & Moe, 2000), which implicates an additional source of negative bias for the income response factor of the EMUC estimator. Informal sector biases might also be present in micro-economic data. 14 Both studies were mainly based on linking together various previous time-use surveys (from which some sources are utilized both in Aguiar and Hurst and in Ramey and Francis). 15 If child care was excluded from the non-market work measure (so that it counts as leisure), the decreases of market work were of the same magnitude as the decreases in household production. However, if child care was included in the non-market work measure, the size of changes in non-market work was around half of the size of changes in market work because child care has increased. 16 Ramey and Francis (2009) claimed that the consensus previous to their study was that there had been a dramatic increase in leisure time. 17 Although fluctuating somewhat in-between the examined time span. 18 Ramey and Francis (2009) briefly commented on the discrepancy in results between the two studies as: “Our results for this population imply somewhat smaller increases in leisure from 1965 to 2003, mainly because we follow the literature and classify child care as home production rather than leisure, we do not count meals at work as “work” in the early sample, and we use CPS data for our


15

Aguiar and Hurst (2007)

Age group (years) 21–65

Increase in leisure (hours/week) 4.6–8.1

Ramey and Francis (2009)

Age group (years) 18–24 25–54 55–64

Increase in leisure (hours/week) 3.66 −2.5 1.12

Table 1: Comparison of results between Aguiar and Hurst (2007) and Ramey and Francis (2009).

3.3 The compensated consumption elasticity w.r.t. leisure

No studies that estimate the compensated consumption elasticity w.r.t. leisure, i.e., 𝜀��𝐿, have been found. When it comes to the compensated consumption elasticity w.r.t. market work, i.e., 𝜀��𝑀, the studies cited in Chetty (2006) showed very small if any signs of any relationship (i.e., they indicated that 𝜀��𝑀 ≈ 0).

However, 𝜀��𝑀 can be used as a proxy of −𝑀

𝐿𝜀��𝐿 only if

𝜕𝐻

𝜕𝑤≈ 0. Otherwise, the

compensated consumption elasticity w.r.t market work needs to be adjusted by the ratio between the response in market work to the response in leisure19, i.e. 𝜕𝑀

𝜕𝑤/ (−

𝜕𝐿

𝜕𝑤), which in turn is hard to find reliable data on.

One crude approach to approximate 𝜀��𝐿 is to assume that consumption responses to income-compensated leisure shocks are proportional to the share of leisure-related expenditures. When using this approach, it is also necessary to estimate what proportion of leisure time gives rise to extra leisure expenditures

and to insert only that part of the leisure into the ratio 𝑀

𝐿 in expression (13) or

(19). Vandenbroucke (2009) reported that the leisure share of expenditures increased from 3.0% in 1900 to 5.8% in 1950 in the United States. Weagley and Huh (2005) reported leisure expenditure shares of 8.1%–8.6 % for near retired and retired people in the United states based on data from 1995.20 In line with these statistics, Pawlowskia & Breuera (2012)21, Thompson and Tinsley (1979), among others concluded that leisure expenditure is mostly a

hours of work measures.” (p. 211). However, it is somewhat unclear what they mean since Aguiar and Hurst used two measures of market work, one that included meals at work and one that did not, and four different measures of leisure, from which two did not include child care. Nevertheless, comparing the estimate results of the studies, it is clear that there is a considerable difference of estimates of core market work around 1965, where Ramey and Francis estimates are lower than Aguiar and Hurst estimates. 19 Remember that

𝑀

𝐿𝜀��𝐿 =

𝜕𝑀

𝜕𝑤/ (

𝜕𝐿

𝜕𝑤) ∙ 𝜀��𝑀 (see Appendix 3).

20 Note that only the consumption bought in the market is included in these studies, but it is preferable that household-produced goods are also included. 21 For Germany, Pawlowskia & Breuera (2012) found that the leisure expenditure share was about 3% in 2006.


16

luxury good, and this might also indicate that complementarity between leisure and consumption might increase with increased income.


17

3.4 Proportion between market work and leisure

The proportion between market work and leisure, i.e. 𝑀

𝐿, during the 20th century

among the age group 25–54 years in the United States, is calculated using Tables

2, 4, and 5 in Ramey & Francis (2009). The mean ratio over the century is 𝑀

𝐿=

0.25 if leisure is calculated by all non-production time (including educational

time, commuting time, and personal care time, e.g. sleep) and 𝑀

𝐿= 0.85 if leisure

is calculated as core (recreational) leisure only.

3.5 Proportion of unearned income

When it comes to the proportion of unearned income to earned income, 𝑦

𝑤𝑀,

Chetty used the ratio between total capital income to total labor income in the United States (equal to 0.5) 22 because he argued that they should be equal on an aggregate level23. He further argued that the skewed distribution of wealth

implies that 𝑦

𝑤𝑀< 0.5 for most households.

However, in the present study, 𝑦

𝑤𝑀 is calculated with more precision (and only

for individuals participating in market work) for two example countries, the United States and Sweden.24 For the United States, the median 𝑦 is divided by median 𝑤𝑀 (data from U.S. Census Bureau) for each sex and year between 1960

and 1968 and restricted to full-time workers. The result is that 𝑦

𝑤𝑀 is in the

range 0.00–0.08, with a tendency for higher values in the latter period25. For Sweden, mean 𝑦 is divided by mean 𝑤𝑀 (data from Statistics Sweden) for five different worker categories and for each year from 2003 to 201326 . For

white and blue-collar workers, 𝑦

𝑤𝑀 ranges from 0.08 to 0.18, and there is a

tendency that 𝑦

𝑤𝑀 is decreasing in salary for a given year. For farmers/self-

employed and other unspecified workers, 𝑦

𝑤𝑀 reaches as high as 0.36.

22 A similar method to estimate

𝑦

𝑤𝑀 is to base it on labor shares, that is, the ratio between aggregate

costs of labor to total GDP. Using the formula 𝑦

𝑤𝑀≈ (1 − 𝑙𝑎𝑏𝑜𝑢𝑟_𝑠ℎ𝑎𝑟𝑒 )/𝑙𝑎𝑏𝑜𝑢𝑟_𝑠ℎ𝑎𝑟𝑒, where

𝑙𝑎𝑏𝑜𝑢𝑟_𝑠ℎ𝑎𝑟𝑒 ranges from about 60% to 80% in developed countries and 50% to 70% in developing

countries based on Serres et al (2002) and Figures 31 and 32 in Global Wage Report 2012/13, 𝑦

𝑤𝑀

seems to be in the range 0.25 to 0.67 in developed countries and in the range 0.43 to 1.0 in developing countries during the years 1970–2010. The United States is of special interest because

many other estimates used in the present study are based on U.S. data. For the United States, 𝑦

𝑤𝑀

based on labor share is about 0.56–0.67. 23 Unearned income for individuals includes transfers, but on an aggregate level these are assumed to cancel so that the only source left is capital income. 24 These countries were chosen due to good availability of official data. The US is important because many other estimates used in the present study are based on US data. 25 Between 1960 and 1968,

𝑦

𝑤𝑀 is in the range 0.00–0.03, and between 1969 and 2014

𝑦

𝑤𝑀 is in the

range 0.01–0.08. 26 Except for two outlier years, 2007 and 2008, after a new government took power.


18

3.6 Summary of the empirical evidence

The empirical evidence from the previous section is summarized in Table 2.

Parameters Source Description Results

Uncompensated labor

supply elasticity

Evers et al (2008) Review, 30 studies,

OECD

−0.24 to 2.8, mostly

>0

Macroeconomic

studies

2 regressions,

1 GE model <0

Descriptive studies of

the development of

division of time

2 studies based on

overlapping survey

studies from the U.S.

Mostly <0

Ramey and Francis

(2009) showed

unchanged time

spent on labor supply

from prime-age

individuals

Changes in household

production and leisure

Aguiar and Hurst

(2007)

Development in the

U.S. 1965–2003,

among prime-age

individuals, based on

multiple survey

studies.

Household

production has

decreased, leisure

has increased

Ramey and Francis

(2009)

Development in the

U.S. 1900–2005,

based on various

data sources, mainly

survey studies.

Household

production has

decreased, leisure

has increased for

total population, no

changes for prime-

age population

Compensated

consumption elasticity

w.r.t. leisure

Chetty (2006) Summary of previous

studies

𝜀𝑐,𝑀𝑐 ≈ 0

Vandenbroucke

(2009),

Weagley and Huh

(2005)

Leisure expenditure

shares in the US

Leisure expenditures

increased from 3% in

1900 to around 8%

in 1995

Proportion between

market work and

leisure

Ramey and Francis

(2009)

1900–2000 for age

group 25–54 years

𝑀

𝐿= 0.85 based on

core leisure only

𝑀

𝐿= 0.25 if leisure =

non-production time

Proportion unearned

income, 𝑦

𝑤𝑀

U.S. Census Bureau Full-time workers, US 0.00–0.08

Statistics Sweden Working population,

Sweden

0.08–0.18 for white

and blue-collar

workers

Table 2: Summary of empirical evidence for relevant parameter values.


19

3.7 Discussion of parameter values

When reviewing the results from the labor supply studies, it seems that the uncompensated labor supply elasticity w.r.t. wage is positive. However, these studies are often contextual, since they do not control for what the alternative to market work is. For example, for an individual who has housekeeping as her main task, and who has a market wage that is close to the price of such services, i.e. 𝑔′′(𝐻) ≈ 0, it is rational to have a high labor supply elasticity w.r.t. wage irrespective of her EMUC value (as indirectly shown in equation (12a)).27 Since the underlying aim of many labor supply studies is to uphold the tax base when reforms are made, it is not a problem that the labor supply elasticity of an individual is high because she has the option to hire staff. But if these contextual values are used to make inference about the utility function, the results will be biased, as shown in Section 2.3.28 It is worth noting that Ramey and Francis’ (2007) results suggest that, for the age group 25–54 years old, it is possible that decreases in working times are largely driven by an increase in women’s participation rate. If the average wage is strongly correlated with women’s participation, then it is possible that estimates of the uncompensated labor supply elasticity from the macroeconomic studies could be negatively biased. However, the long-run historical trend is not an increase of mean working time, even for the prime-age

population, which suggests the relationship 𝜕𝑀

𝜕𝑤≤ 0.

The analysis in Section 2.4 shows that in order to calculate a point estimate of EMUC based on labor supply responses to wage changes, household production responses need to be estimated, which is not necessary for a lower bound according to Section 2.3. However, it is difficult to draw conclusions on utility parameters from changes over long time periods as there may have been structural changes, for instance, due to changes in household production technology, changes in social norms towards division of labor within the household, etc. In other words, the function 𝑔(𝐻) is likely to have changed somewhat. This means that the point estimates are much less reliable than the lower bound estimates.29 Ramey and Francis found that leisure time has increased, whereas both household production and market work have decreased for the total population, while time allocation has remained stable in the long run for the prime-age population. The implied difference in elasticities between the total population and the prime-age population relates to partly different questions. The

27 This scenario; i.e. 𝑔′′(𝐻) ≈ 0, could explain why the extensive margin responses in labor supply are larger than the intensive margin responses (see equation (12a)). 28 Also note that since both the models in the present study and in Chetty’s (2006) use a linear budget restriction, it might be that these models are somewhat unsuitable to estimate with individual data in cases where the individuals face non-linear budget restriction due to kinks in the income tax scheme. 29 In principle, changes in productivity of household production is also relevant for the lower bound, but only if household production productivity increases are of at least the same magnitude as the general productivity increase in the market-based economy, which I find unrealistic.


20

estimates for the age group 25–54 years old relate to within-time-period optimization of leisure and consumption, whereas the estimates from the total population roughly relate to lifetime optimization. Because it is not straightforward which one of them to choose, both approaches will be considered in Section 4 (estimation results).

For the proportion between market work and leisure, i.e. 𝑀

𝐿, one might think that

it makes a huge difference if 𝑀

𝐿= 0.25 or

𝑀

𝐿= 0.85. Fortunately, however, the

proposed EMUC formulas are not very sensitive to 𝑀

𝐿 when the complementarity

between leisure and consumption is modest. For this reason, I will use 𝑀

𝐿= 0.85,

based on core leisure only, for the main analysis. The rationale for this is that the share of leisure-related expenditures is used as a proxy for the complementarity between leisure and consumption30, and it is reasonable that only core leisure time influences the demand for leisure-related expenditures.

When it comes to the proportion of unearned income, 𝑦

𝑤𝑀, there is a large

difference between proxy estimates based on aggregate data and the more detailed estimates based on more disaggregate data for the working population, especially for the US case.31 The Swedish numbers for white and blue-collar workers range from 0.08 to 0.18, which is still notably higher than the US range of 0.00–0.08. One reason can be that Swedish data are not restricted to full-time workers. Another probable explanation is that Sweden has a general welfare system that includes a general monthly child support of about 100 euros per child32. As a conclusion of this section, evidence from historical macroeconomic data suggests that uncompensated labor supply elasticity w.r.t wage is negative, i.e., 𝜕𝑀

𝜕𝑤≤ 0. Evidence for the development of leisure time and household production

is inconclusive, but none of the studies cited indicate that time spent on household production has increased. The leisure expenditure share of consumption has increased from about 0.03 at the beginning of the 20th century to about 0.08 at the end of the century in the United States. The mean ratio of market work to leisure has been about 0.85 when only core leisure is included, and the ratio of unearned income to earned income has ranged from about 0.00 to 0.18 for the working population in the United States and Sweden during the

30 In turn, the rationale for using leisure-related expenditures as a proxy for complementarity between leisure and consumption is the assumption that sleep does not generate extra expenses and that the cost of commuting, personal care, and education are fairly inelastic w.r.t. non-production time, or at least marginal in size on average. 31 For the Swedish case, some of the highest estimates based on disaggregate data are in line with the lowest estimates based on aggregate data for developing countries. On the other hand, the highest estimates from the detailed data are from self-employed/farmers and other unspecified workers. When it comes to being self-employed, the division between labor income and capital income is not straightforward, and because it is more favorable from a tax avoidance perspective to declare income as capital income, these figures may not be reliable. However, numbers for white and blue-collar workers can be considered more reliable because they are based on third-party reporting (employers, banks, etc.). 32 Possibly in combination with a less skewed capital distribution.


21

last decades. Based on these stylized facts, it is possible to derive estimates of EMUC, as will be done in the following section.

4 RESULTS In this section I present numerical estimation results of EMUC – first in the form of a sensitivity analysis of Chetty’s results, and second in the form of point estimates and lower bounds based on the new model that includes household production.

4.1 Sensitivity analysis of Chetty’s results

In this section I perform a sensitivity analysis of Chetty’s (2006) result, starting with the bias given by the complementarity error factor (equation (16c)). I will also look at the sensitivity to unearned income.

In Table 3, the values of the complementarity error factor, ( 𝜂

𝜂

)𝐵

=

1−(1+𝑦


𝜕𝑀

𝜕𝑤/(−

𝜕𝐿

𝜕𝑤)

1−(1+𝑦


, for different values of the input parameters 𝜕𝑀

𝜕𝑤/ (−

𝜕𝐿

𝜕𝑤),

𝑦

𝑤𝑀, and 𝜀��𝑀 are presented.

Response proportions

(x units) 𝝏𝑴

𝝏𝒘/ (−

𝝏𝑳

𝝏𝒘)

Complementarity error factor,( 𝜼

𝜼

)𝑩

𝒚

𝒘𝑴 0 0.5

𝝏𝑴

𝝏𝒘

𝝏𝑯

𝝏𝒘

𝝏𝑳

𝝏𝒘 𝜺𝒄,𝑴

𝒄 −0.1 0.1 −0.01 0.01

1 −1 +0 −∞ −∞ +∞ −∞ +∞

1 −1.015 0.015 ~−66.67 −5.15 8.52 0.00 2.03

1 −1.1 0.1 −10 0.00 2.22 0.84 1.17

1 −2 1 −1 0.82 1.22 0.97 1.03

0 −1 1 0

0.91 1.11 0.99 1.02

−1 −1 2 0.5 0.95 1.06 0.99 1.01

1 0 −1 1

1.00 1.00 1.00 1.00

1 −0.5 −0.5 2

1.09 0.89 1.01 0.98

1 ~−0.67 ~−0.33 3 1.18 0.78 1.03 0.97

1 −0.9 −0.1 10

1.82 0.00 1.13 0.86

1 −0.985 −0.015 ~66.67

6.97 −6.30 1.97 0.00

1 −1 −0 +∞ +∞ −∞ +∞ −∞

Table 3: The value of the complementarity error factor,( 𝜂

𝜂

)𝐵

, for different parameter values.

The first three columns show some (plausible) proportions in time allocation

responses to a wage change that result in a particular value of 𝜕𝑀

𝜕𝑤/ (−

𝜕𝐿

𝜕𝑤),


22

which in turn will determine the value of the complementarity error factor33,

( 𝜂

𝜂

)𝐵

. This error factor is calculated for two type cases. One is a reference case

where 𝜀��𝑀 = ±0.1 and there is no unearned income. The other type case is constructed to match the findings in Chetty (2006) that 𝜀��𝑀 is small in absolute terms and unearned income is half that of earned income. We now turn to sensitivity of Chetty’s results with respect to the proportion of unearned income. When EMUC is estimated from the ratio between the elasticity of production time w.r.t. unearned income and the compensated production time elasticity w.r.t. wage, the formula includes the unearned

income factor (1 +𝑤𝑀

𝑦). The occurrence of (1 +

𝑤𝑀

𝑦) is irrespective of whether

household production is included in the model or not, that is, (1 +𝑤𝑀

𝑦) is

included both in Chetty’s formula (5) and in equation (13), but equation (13) is not used for empirical estimation in the present study34. In Figure 2, the

unearned income factor (1 +𝑤𝑀

𝑦) is plotted as a function of the ratio between

unearned and earned income, 𝑦

𝑤𝑀.

Figure 2: The unearned income factor (1 +

𝑤𝑀

𝑦) of EMUC, included in equations (5) and (13)

(vertical axis), as a function of 𝑦

𝑤𝑀 (horizontal axis).

33 Please keep in mind that (

𝜂

𝜂

)𝐵

= 1 means no bias.

34 The formulas used for empirical estimations in the present study are based on the uncompensated

elasticity of production time, and in these cases the ratio of 𝑤𝑀

𝑦 is reversed so that the unearned

income factor instead is (1 +𝑦

𝑤𝑀) and is thus not as sensitive to small values of

𝑦

𝑤𝑀.

0,03; 34,3

0,08; 13,5

0,18; 6,60,5; 3 1; 2

0

10

20

30

40

50

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

(1+𝑤𝑀/𝑦)

𝑦/𝑤𝑀


23

4.2 Empirical estimates of EMUC, including household production

Based on the broad picture that emerges from the empirical evidence, it is possible to calculate lower bound and point estimates of EMUC using the equations derived in Sections 2.3 and 2.4. Point estimates for the special case that total production time is insensitive to

wage changes, i.e., 𝜕𝑃

𝜕𝑤= 0, coincide with lower bound estimates of EMUC when

total production time decreases with wage, i.e., 𝜕𝑃

𝜕𝑤≤ 0, or when the more

restrictive assumption that also market work decreases with wage holds, i.e., 𝜕𝑀

𝜕𝑤≤ 0. It is assumed that the proportion between market work and leisure is

𝑀

𝐿= 0.85 based on estimates of market work and core leisure from Ramey and

Francis (2009).35 These estimates are calculated using Proposition 4 with different levels of compensated consumption elasticity w.r.t. leisure, 𝜀��𝐿, and different proportions of unearned income, 𝑦/(𝑤𝑀). The results are shown in Table 4:

𝜺𝒄𝒄,𝑳 𝒚/(𝒘𝑴) 0.00 0.03 0.08 0.18

0.00 1.00 1.03 1.08 1.18

0.05 0.96 0.99 1.03 1.12

0.10 0.92 0.95 0.99 1.07

0.15 0.89 0.91 0.95 1.03

Table 4: EMUC estimates for the special case 𝜕𝑃

𝜕𝑤= 0 based on Proposition 4.

That 𝜕𝑃

𝜕𝑤= 0 is suggested by the leisure development for the age group 25–54

years in the United States during the 20th century, according to Ramey and Francis (2009). However, this is not supported by the parallel trend for the total population. The columns denote the proportion of unearned income, where 0.00, 0.03, and 0.08 are based on the median income of full-time workers in the United States over the years 1960–2014 (0.00 is the minimum value, 0.03 is the average, and 0.08 is the maximum) and 0.08 and 0.18 denote the lowest and highest mean values, respectively, for white and blue-collar workers in Sweden between 2003 and 2013. Values of 𝑀/𝐿 are based on the age group 25–54 years in the United States between 1900 and 2005 in Ramey and Francis (2009). The values of 𝜀��𝐿 are loosely related to observed shares of leisure-related expenditures in the United States during the 20th century36.

35 The effect of

𝑀

𝐿= 0.25 and

𝑀

𝐿= 1 has also been examined, and the differences in results are

marginal. For 𝑀

𝐿= 0.25, EMUC estimates are about 0.03 to 0.10 units higher, and thus do not

influence the lower bound. For 𝑀

𝐿= 1 (which represents the extreme value in the data set), EMUC

estimates are 0.01 to 0.03 units lower. Also note that when 𝜀��𝐿 = 0, the value of 𝑀

𝐿 does not

influence the value of the EMUC estimates. 36 The maximum value was 0.08 in 1996. But by including somewhat higher values, I allow for the likely event that the share has continued to grow, as well as adding some safety margins since the link to complementary is uncertain. Also, 𝜀��𝑀 = 0.15 was the maximum value used in Chetty (2006), so including 𝜀��𝐿 = 0.15 facilitates comparison between studies.


24

Table 5 shows point estimates of 𝜂 for the special case of additive utilities

between leisure and consumption, i.e. 𝜕2𝑢

𝜕𝑐𝜕𝐿= 0, based on Proposition 5.

𝝏𝑯

𝝏𝒘/

𝝏𝑴

𝝏𝒘 𝜺𝑴,𝒘 𝒚/(𝒘𝑴) 0.00 0.03 0.08

0

−0.106 1.12 1.15 1.21

−0.124 1.14 1.18 1.23

−0.160 1.19 1.23 1.29

0.5

−0.106 1.19 1.22 1.28

−0.124 1.23 1.27 1.33

−0.160 1.32 1.36 1.42

1

−0.106 1.27 1.31 1.37

−0.124 1.33 1.37 1.44

−0.160 1.47 1.51 1.59

Table 5: Estimates of 𝜂 for the special case 𝜕2𝑢

𝜕𝑐𝜕𝐿= 0, restricted to observed values of

uncompensated labor supply w.r.t. wage, i.e. 𝜀𝑀𝑤, and the ratio between unearned and earned

income, 𝑦

𝑤𝑀, based on Proposition 5.

Estimates of 𝜀𝑀,𝑤 are from Huberman and Minns (2007) and are based on firm-

level data for 14 OECD countries, predominantly the United States, in the period 1870–1900 (restricted to full sample estimates with different model specifications37). The proportion of unearned income is based on median values of earned and unearned income for each sex and year between 1960 and 2014

in the United States38. 𝑦

𝑤𝑀= 0 denotes the sample minimum (women in 1965),

0.03 is the sample mean over the years (same for both sexes), and 0.08 is the

sample maximum (women in 2006). 𝜕𝐻

𝜕𝑤/

𝜕𝑀

𝜕𝑤= 0 is not an empirical value, but it

is included for comparison. The other two figures of 𝜕𝐻

𝜕𝑤/

𝜕𝑀

𝜕𝑤 correspond roughly

to the parallel trends in household production and market work during 1965–2003 in the United States, controlled for demographic changes, given by Aguiar

and Hurst (2007). 𝜕𝐻

𝜕𝑤/

𝜕𝑀

𝜕𝑤= 0.5 corresponds to the case that child care is

defined as household production, and 𝜕𝐻

𝜕𝑤/

𝜕𝑀

𝜕𝑤= 1 corresponds to the case

where child care is defined as leisure.

37 The values of 𝜀𝑀,𝑤 are based on the parameter estimates for log wage in Tables 4 and 5 in

Huberman and Minns (2007) where −0.106 is the minimum value among 8 different model specifications, −0.124 is the mean value, and −0.160 is the maximum value. Note that the minimum of −0.106 denotes the specification that includes the most control variables, all of which are significant. 38 Since 𝜀𝑀,𝑤 are old and based on data mostly from the United States, the more recent Swedish

estimates of 𝑦

𝑤𝑀 do not seem relevant.


25

5 DISCUSSION

In Table 3, the value of the complementarity error factor, ( 𝜂

𝜂

)𝐵

=

1−(1+𝑦


𝜕𝑀

𝜕𝑤/(−

𝜕𝐿

𝜕𝑤)

1−(1+𝑦


, is estimated. Note that when the uncompensated labor

supply elasticity w.r.t wage is negative, i.e., 𝜕𝑀

𝜕𝑤< 0, then the following

relationship holds: 𝜕𝑀

𝜕𝑤/ (−

𝜕𝐿

𝜕𝑤) ∈ [0,1]. This means that the size of the error

from ( 𝜂

𝜂

)𝐵

will be less than 11% when 𝜀��𝑀 < 0.1 and there is no unearned

income, and it will be less than 2% when 𝜀��𝑀 < 0.01 and unearned income is half of earned income.

On the other hand, among the studies cited by Chetty, 𝜕𝑀

𝜕𝑤 was mostly positive

and the values of 𝜕𝐿

𝜕𝑤 were not known. Table 3 and Figure 2 show that (

𝜂

𝜂

)𝐵

is

well behaved close to 𝜕𝑀

𝜕𝑤/ (−

𝜕𝐿

𝜕𝑤) = 1, but when |

𝜕𝐿

𝜕𝑤| is small compared to |

𝜕𝑀

𝜕𝑤|,

the function ( 𝜂

𝜂

)𝐵

becomes unstable, ranging from 0 to ±∞. From this

perspective it is worrisome that there is some indication that 𝜕𝐿

𝜕𝑤 might be close

to 0 (Ramey and Francis (2009)). In addition, theoretical analysis (equation (17b)) shows that the income

response error factor tends to zero as |𝜕𝐻

𝜕𝑤| approaches

𝜕𝑀𝑐

𝜕𝑤 , which means that

(𝜂

𝜂

)𝐴

is also sensitive to large values of |𝜕𝐻

𝜕𝑤|. Lastly, I showed in Figure 3 that

Chetty’s results are sensitive to small values of the share of unearned income, which are supported by empirical data. This sensitivity is not exclusive to Chetty’s model, but occurs because EMUC is estimated from the ratio between the uncompensated labor supply elasticity w.r.t. wage and the labor supply elasticity w.r.t. unearned income (instead of only the uncompensated labor elasticity w.r.t. wage). From this perspective, it is recommended that formula (13) be directly estimated only with great caution and only if reliable estimates of the proportion of unearned income are available with high precision. Unfortunately, using a simplified39 version of Proposition 4 (as in the present study) Chetty (2006, p. 1829) gives the impression that the EMUC estimates in formula (5) are also decreasing along with the proportion of unearned income, when in fact the opposite relationship holds. In principle, this undermines the upper bounds of EMUC estimated by equation (9) in that study, since Chetty

does not consider any proportions of unearned income other than 𝑦

𝑤𝑀= 0.5.

Let us now turn to the new estimates of EMUC from the present study. The results presented in Table 4 suggest that EMUC is higher than 0.9. In the two

39Chetty’s (2006) version of the inequality is

𝜕𝑀

𝜕𝑤≥ 0

⇔ �� < 1 +

𝑦

𝑤𝑀, that is, household production

and complementarity, are left out compared to Proposition 4.


26

special cases of 𝜕𝑃

𝜕𝑤= 0 and

𝜕2𝑈

𝜕𝑐𝜕𝐿= 0, EMUC is estimated to be in the range 0.9–

1.6. None of these results contradict Chetty’s main empirical findings that EMUC is bound below 2. However, the average estimate of that study for the case when there is no complementarity between labor supply and consumption was 0.7140, which is below the lowest bound estimated here. This estimate was driven down by a predominance of micro-data studies. The only two macro studies41 cited in Chetty (2006) yielded significantly higher estimates (1.74 and 1.78), which are higher than all the estimates in the present study. Also, had Chetty

used, for example, 𝑦

𝑤𝑀= 0.03 instead of

𝑦

𝑤𝑀= 0.5, all results of that study would

have been about 11 times higher, as can be seen in Figure 2. Let us now turn to the new estimate results from the present study. There are two obvious problems with the estimates in Table 5. The first is that that the wage elasticity of labor supply used is derived from old data42, which unfortunately means that they could be biased due to diminishing productivity of work hours. The second is that the observed estimates of the other included parameters are more recent. Interestingly, EMUC estimates in Table 5 correspond well to EMUC estimates in previous studies (Evans, 2005). It is worth noting that Table 4 suggests that the lower bound of EMUC is close to unity, which is the central value used in several important recent CBA applications, for example, The Stern Review (Stern, 2006) and the UK Treasury (Evans, 2006).

6 CONCLUSION In the present study, I have shown theoretically that the results in the original study by Chetty (2006) are sensitive to the proportion of unearned income and that the omission of household production may lead to bias. In particular, if the observed compensated labor supply responses are partly counteracted by unobserved changes in household work, and if the compensated consumption elasticity w.r.t. leisure is negligible (as was suggested in Chetty’s study), this model will produce a negative bias, which in turn will compromise the upper

40 The mean estimate of 0.71 is low compared to the EMUC estimates in former studies (Evans, 2005), but Chetty did not at all relate to the previous EMUC literature. His aim was instead to show that his EMUC estimates did not support an expected utility-based formulation of risk aversion. 41 However, these two studies suffered from multiple problems; the most serious is that they did not look at responses to wage changes, but to income tax changes. On a macroeconomic level, tax changes do not only influence labor supply responses through net wage, but again, through division of work where higher taxes will lead to a larger degree of unpaid work and a larger informal sector. In addition, the two studies were not peer reviewed and were based on partly the same data. 42 One positive aspect of this is that complementarity between leisure and consumption was probably negligible at this time (leisure expenditure as the share of consumption was low around the year 1900 according to Vandenbroucke (2009)), which implies that 𝜀𝑐,𝐿

𝑐 ≈ 0 is a reasonable

approximation.


27

bound43. This happens because substitution between consumption and leisure cannot be distinguished from substitution between market work and household production, which is a choice independent of EMUC. I have also derived point estimators of EMUC for two plausible special cases from which one can be used to estimate a lower bound of EMUC under more general conditions. I further have offered some empirical estimates based on current evidence of the included parameters where the lower bound of EMUC is about 0.9 and point estimates are in the range 0.9–1.6. An EMUC value larger than unity means that the marginal utility of consumption diminishes fast enough for extra leisure to be more attractive than extra income when wages increase. Even though a person who gets a raise can consume more, she still prefers to work less because the extra leisure is worth more at the margin. The present study has suggested a lower bound of EMUC close to unity, which is substantially higher than some of the previous estimates (including the mean estimate in Chetty (2006)). The mechanism behind this result is that extra leisure can be earned without reducing the labor supply. The reason is that a wage increase can be spent in order to devote less time to household work. Obvious examples of options for substituting consumption for time spent in household production include purchasing a washing machine or a dishwasher. This means that an individual, whose EMUC value is higher than unity, might still choose not to reduce her working time as her wage increases. This is the basic logic for motivating why the value of EMUC should be higher than indicated by Chetty (2006). This may have important consequences for the appropriate way to trade off current and future consumption, as a higher EMUC value implicates a higher discount rate as well as increased impact of risk aversion and equity weighting when applicable. One difficulty in estimating EMUC using the models developed in the present study has been to find unbiased estimates of the uncompensated labor supply elasticity w.r.t. wage. In order to obtain reliable point estimates of EMUC from the proposed formulas, new research with the aim to study the effect of wages on total work supply, rather than the effect of taxes on individual labor supply, is ultimately needed. As a last point, I want to address the possibility that the within-household substitution between paid and unpaid work might be of increasing importance due to increasing gender equality. As gender roles become less rigid, more tasks are socially acceptable to handle for both sexes. As more women participate in market work, the substitution between housework and market work becomes more explicit (if a woman stays at home full time, it might be natural to handle the majority of housework independent of her husband’s wage). The implication is that bias from a model without household production, estimated with individual data, might increase over time.

43 The upper bound of 2 is also not strongly supported by macroeconomic data; the two macroeconomic studies included in Chetty’s study both yielded upper bound estimates of 2.3.


28

APPENDIX 1 Setting 𝐿 = 𝜏 − 𝑀 − 𝐻, (8) can be rewritten as a maximization problem in 𝑀, 𝐻, with the FOC:

𝜕𝑢

𝜕𝑐∙ 𝑤 −

𝜕𝑢

𝜕𝐿= 0, (A1.1)

𝜕𝑢

𝜕𝑐∙ 𝑔′(𝐻) −

𝜕𝑢

𝜕𝐿= 0. (A1.2)

It follows that: 𝑔′(𝐻) = 𝑤. (A1.3) From this, we see that, for an internal solution, 𝐻 is dependent only on 𝑤 and

hence independent of all other variables, for example 𝑦 or 𝑀. Thus, 𝜕𝐻

𝜕𝑦= 0,

which in turns implies:

𝜕𝐿

𝜕𝑦= −

𝜕𝑀

𝜕𝑦. (A1.4)

Differentiating (A1.3) w.r.t. 𝑤 gives:

𝑔′′(𝐻) ∙𝜕𝐻

𝜕𝑤= 1,

𝜕𝐻

𝜕𝑤=

1

𝑔′′(𝐻). (A1.5)

Thus, 𝑔′′(𝐻) < 0 (from equation (8)) gives:

𝜕𝐻

𝜕𝑤< 0. (A1.6)

APPENDIX 2 (A1.1) is basically the same equation as equation (3) in Chetty (2006). It allows us to follow the procedure given by equation (4) in Chetty (2006): Differentiation of (A1.1) w.r.t. 𝑦:

𝑤𝜕2𝑢

𝜕2𝑐(1 + 𝑤 ∙

𝜕𝑀

𝜕𝑦) + 𝑤

𝜕2𝑢

𝜕𝑐𝜕𝐿∙

𝜕𝐿

𝜕𝑦=

𝜕2𝑢

𝜕2𝐿∙

𝜕𝐿

𝜕𝑦+

𝜕2𝑢

𝜕𝑐𝜕𝐿∙ (1 + 𝑤 ∙

𝜕𝑀

𝜕𝑦). (A2.1)

Inserting 𝜕𝐿

𝜕𝑦= −

𝜕𝑀

𝜕𝑦 (from A1.4) and solving for

𝜕𝑀

𝜕𝑦 gives:

𝜕𝑀

𝜕𝑦=

−𝑤𝜕2𝑢

𝜕2𝑐+

𝜕2𝑢

𝜕𝑐𝜕𝐿

𝑤2𝜕2𝑢

𝜕2𝑐+

𝜕2𝑢

𝜕2𝐿−2𝑤

𝜕2𝑢

𝜕𝑐𝜕𝐿

. (A2.2)

Differentiation of (A1.1) w.r.t. 𝑤:

𝜕𝑢

𝜕𝑐+

𝜕2𝑢

𝜕2𝑐(𝑤 ∙

𝜕𝑀

𝜕𝑤+ 𝑀 + 𝑔′(𝐻) ∙

𝜕𝐻

𝜕𝑤) ∙ 𝑤 +

𝜕2𝑢

𝜕𝑐𝜕𝐿∙

𝜕𝐿

𝜕𝑤∙ 𝑤 =

=𝜕2𝑢

𝜕2𝐿∙

𝜕𝐿

𝜕𝑤+

𝜕2𝑢

𝜕𝑐𝜕𝐿∙ (𝑤 ∙

𝜕𝑀

𝜕𝑤+ 𝑀 + 𝑔′(𝐻) ∙

𝜕𝐻

𝜕𝑤), (A2.3)


29

In the following, it will be convenient to define total production time as: 𝑃 = 𝑀 + 𝐻 . (A2.4) Which means that:

𝜕𝑃

𝜕𝑤=

𝜕𝑀

𝜕𝑤+

𝜕𝐻

𝜕𝑤. (A2.5)

Which means that the time constraint in equation (8) gives:

𝜕𝐿

𝜕𝑤= −

𝜕𝑃

𝜕𝑤. (A2.6)

Inserting (A2.5) and (A2.6) into (A2.3) and solving for 𝜕𝑃

𝜕𝑤 gives:

𝜕𝑃

𝜕𝑤=

−𝜕𝑢

𝜕𝑐−𝑤𝑀

𝜕2𝑢

𝜕2𝑐+𝑀

𝜕2𝑢

𝜕𝑐𝜕𝐿

𝑤2𝜕2𝑢

𝜕2𝑐+

𝜕2𝑢

𝜕2𝐿−2𝑤

𝜕2𝑢

𝜕𝑐𝜕𝐿

. (A2.7)

Following the procedure given by equations (5) and (6) in Chetty (2006), 𝜂 can be solved from the expression (after simplification):

𝜕𝑀

𝜕𝑦

𝜕𝑃

𝜕𝑤−𝑀

𝜕𝑀

𝜕𝑦

=−𝑤

𝜕2𝑢

𝜕2𝑐+

𝜕2𝑢

𝜕𝑐𝜕𝐿

−𝜕𝑢

𝜕𝑐

. (A2.8)

𝜕2𝑈

𝜕𝑐𝜕𝐿 can be solved for in (A2.8), by following the procedure prior to equation (8)

in Chetty (2006): Consider an agent who faces the possibility of two states 1 and 2 with probability p of state 1. Consumption can be traded freely between states at a rate determined by p. The agent faces the following maximization problem: max

𝑐1,𝑐2

(𝑝 ∙ 𝑢(𝑐1, 𝐿1) + (1 − 𝑝) ∙ 𝑢(𝑐2, 𝐿2)), (A2.9)

s.t. 𝑝 ∙ 𝑐1 + (1 − 𝑝) ∙ 𝑐2 = 𝑝 ∙ 𝑌1 + (1 − 𝑝) ∙ 𝑌2, where 𝑌𝑠, is the total income in each state, S: 𝑌𝑠 = 𝑦𝑠 + 𝑤𝑠𝑀𝑠 + 𝑔(𝐻𝑠), where 𝐻𝑠 is the optimal level of household production time in each state and 𝑀𝑠 can be either exogenous or set at an optimal level. Hence, 𝑌𝑠 can be treated as exogenous when optimizing the level of consumption in each state. It follows that:

𝜕𝑢(𝑐1,𝐿1)

𝜕𝑐=

𝜕𝑢(𝑐2,𝐿2)

𝜕𝑐. (A2.10)


30

A first-order Taylor expansion 𝜕𝑢

𝜕𝑐 around 𝑐1 gives:

𝜕𝑢(𝑐2,𝐿2)

𝜕𝑐=

𝜕𝑢(𝑐1,𝐿1)

𝜕𝑐+

𝜕2𝑢(𝑐1,𝐿1)

𝜕2𝑐∙ ∆𝑐 +

𝜕2𝑢(𝑐1,𝐿1)

𝜕𝑐𝜕𝐿∙ ∆𝐿 + 𝑅, (A2.11)

where ∆𝑐 = 𝑐2 − 𝑐1 and ∆𝐿 = 𝐿2 − 𝐿1. Inserting (A2.10) into (A2.11) gives:

𝜕2𝑢(𝑐1,𝐿1)

𝜕𝑐𝜕𝐿∆𝐿 =

−𝜕2𝑢(𝑐1,𝐿1)

𝜕2𝑐∆𝑐 − 𝑅, (A2.12)

𝜕2𝑢(𝑐1,𝐿1)

𝜕𝑐𝜕𝐿= lim

∆𝐿→0(−

∆𝑐

∆𝐿∙

𝜕2𝑢(𝑐1,𝐿1)

𝜕2𝑐), (A2.13)

𝜕2𝑢

𝜕𝑐𝜕𝐿= −

𝜕𝑐𝑐

𝜕𝐿∙

𝜕2𝑢

𝜕2𝑐. (A2.14)

where 𝜕𝑐𝑐

𝜕𝐿 is the response in consumption due to a marginal and fully

compensated change in leisure.

Inserting (A2.14) into (A2.8) and multiplying by – 𝑐 gives:

−𝑐∙

𝜕2𝑢

𝜕2𝑐𝜕𝑢

𝜕𝑐

∙ (𝑤 +𝜕𝑐𝑐

𝜕𝐿) = −

𝑐∙𝜕𝑀

𝜕𝑦

𝜕𝑃𝑐

𝜕𝑤

, (A2.15)

𝜂 = −

𝜕𝑀

𝜕𝑦

𝜕𝑃𝑐

𝜕𝑤

∙𝑐

𝑤+𝜕𝑐𝑐

𝜕𝐿

, (A2.16)

𝜂 = −

𝜕𝑀

𝜕𝑦∙

𝑦

𝑀

𝑃

𝑀∙𝜕𝑃𝑐

𝜕𝑤∙𝑤

𝑃

∙

𝑐

𝑦

1+𝑐

𝑤𝑀∙𝑀

𝐿∙𝜕𝑐𝑐

𝜕𝐿∙𝐿

𝑐

, (A2.17)

𝜂 = −𝜀𝑀𝑦

𝑃

𝑀��𝑃𝑤

∙(1+

𝑤𝑀

𝑦)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿∙��𝑐𝐿)

. (A2.18)


31

APPENDIX 3 (A2.13) can be expressed as:

𝜂 = −𝑀

𝜕𝑀

𝜕𝑦

𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐

𝜕𝑤

∙(1+

𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

. (A3.1)

The corresponding equation to (A2.13) using Chetty’s original model is equation (5), which can be expressed as:

�� = −𝑀

𝜕𝑀

𝜕𝑦

𝜕𝑀𝑐

𝜕𝑤

∙(1+

𝑦

𝑤𝑀)

(1−(1+𝑦


, (A3.2)

which means that:

𝜂

𝜂

=𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐


𝜕𝑤

∙1+(1+

𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿

1−(1+𝑦


. (A3.3)

Inserting:

𝜀��𝐿 =𝜕𝑐𝑐

𝜕𝐿∙

𝐿

𝑐=

𝜕𝑐𝑐

𝜕𝑀∙

𝜕𝑀

𝜕𝐿∙

𝐿

𝑐, (A3.4)

gives that:

𝜂

𝜂

= (1 +𝜕𝐻


𝜕𝑤

) ∙1+(1+

𝑦

𝑤𝑀)

𝜕𝑐𝑐

𝜕𝑀∙𝑀

𝑐∙𝜕𝑀

𝜕𝐿

1−(1+𝑦


, (A3.5)

which can be expressed as:

𝜂

𝜂

= (1 +𝜕𝐻


𝜕𝑤

) ∙

1−(1+𝑦


𝜕𝑀

𝜕𝑤/(−

𝜕𝐿

𝜕𝑤)

1−(1+𝑦


. (A3.6)

Let us now turn to (𝜂

𝜂

)𝐴

= 1 +𝜕𝐻


𝜕𝑤

from equation (16a). Because 𝜕𝑀𝑐

𝜕𝑤> 0 and

𝜕𝐻

𝜕𝑤< 0, (

𝜂

𝜂

)𝐴

∈ (0,1] holds as long as 𝜕𝑀𝑐

𝜕𝑤> |

𝜕𝐻

𝜕𝑤|. The question that remains to be

answered is if it is possible that 𝜕𝑀𝑐

𝜕𝑤≤ |

𝜕𝐻

𝜕𝑤|. To answer that question, let us

define the compensated total work supply as:

𝜕𝑃

𝜕𝑤

𝑐=

𝜕𝑃

𝜕𝑤− 𝑃

𝜕𝑃

𝑦. (A3.7)

Using simple addition and subtraction, it is straightforward to show that:

𝜕𝑃

𝜕𝑤

𝑐=

𝜕𝑀𝑐

𝜕𝑤+

𝜕𝐻

𝜕𝑤. (A3.8)


32

Which means that:

𝜕𝑀𝑐

𝜕𝑤≤ |

𝜕𝐻

𝜕𝑤|

⇔

𝜕𝑃𝑐

𝜕𝑤≤ 0. (A3.9)

Expanding (A3.8) gives:

𝜕𝑃𝑐

𝜕𝑤=

𝜕𝑃

𝜕𝑤− 𝑀

𝜕𝑀

𝜕𝑦=

−𝜕𝑢

𝜕𝑐

𝑤2𝜕2𝑢

𝜕2𝑐+

𝜕2𝑢

𝜕2𝐿−2𝑤

𝜕2𝑢

𝜕𝑐𝜕𝐿

. (A3.10)

From (A3.10) it follows that 𝜕𝑃𝑐

𝜕𝑤= 0 can only occur at the asymptotic limit,

lim𝑐→∞

(𝜕𝑢

𝜕𝑐) = 0 (and only if the numerator of (A3.10) approaches zero faster than

the denominator with increasing consumption). 𝜕𝑃𝑐

𝜕𝑤= 0 will therefore be

disregarded in the following. It also follows from (A3.10) that:

𝜕𝑃𝑐

𝜕𝑤< 0

⇔ 𝑤2 𝜕2𝑢

𝜕2𝑐+

𝜕2𝑢

𝜕2𝐿− 2𝑤

𝜕2𝑢

𝜕𝑐𝜕𝐿> 0, (A3.11)

𝜕2𝑢

𝜕𝑐𝜕𝐿<

1

2(𝑤

𝜕2𝑢

𝜕2𝑐+

1

𝑤

𝜕2𝑢

𝜕2𝐿). (A3.12)

where the right hand side of (A3.12) is negative. The economic interpretation is that there needs to be a large degree of substitutability between 𝑐 and 𝐿. In the limiting case with perfect substitutability between 𝑐 and 𝐿, we have: 𝑢(𝑐, 𝐿) = 𝑓(𝑘𝑐 + 𝐿), (A3.13) where 𝑘 > 0. The second-order conditions are:

𝜕2𝑢

𝜕2𝑐= 𝑘2𝑓′′,

(A3.14)

𝜕2𝑢

𝜕𝑐𝜕𝐿= 𝑘𝑓′′,

𝜕2𝑢

𝜕2𝐿= 𝑓′′,

where 𝑓′′ is negative.

Then, if 𝜕𝑃𝑐

𝜕𝑤< 0, (A3.13) inserted into (A3.10) gives:

𝑤2𝑘2𝑓′′ − 2𝑤𝑘𝑓′′ + 𝑓′′ > 0, 𝑤2𝑘2 − 2𝑤𝑘 + 1 < 0. (A3.15)


33

which does not have any real roots for 𝑤𝑘, which indicates that (A3:12) would represent an even higher degree of substitutability than for perfect substitutes. This in turn would implicate a concave indifference curve (see Figure 3).

Figure 3: Two different indifference curves. Case A represents a normal, convex indifference curve, and Case B represents a concave indifference curve.

From Figure 3 it is easy to see that when the indifference curve is concave, the only two optimal mixes of 𝑐 and 𝐿 are the two extremes of 100% consumption or 100% leisure, which violates either equation (9a) or (9b)44.

As a conclusion 𝜕𝑃𝑐

𝜕𝑤> 0 must hold, and hence:

𝜕𝑀𝑐

𝜕𝑤> |

𝜕𝐻

𝜕𝑤|, (A3.16)

(𝜂

𝜂

)𝐴

∈ (0,1]. (A3.17)

For (𝜂

𝜂

)𝐵

we have that:

|(𝜂

𝜂

)𝐵

− 1| = ||

1−(1+𝑦


𝜕𝑀𝜕𝑤

(−𝜕𝐿𝜕𝑤

)−(1−(1+

𝑦


1−(1+𝑦


||,

|(𝜂

𝜂

)𝐵

− 1| = ||

(1+𝑦

𝑤𝑀)��𝑐𝑀−(1+

𝑦


𝜕𝑀𝜕𝑤

(−𝜕𝐿𝜕𝑤

)

1−(1+𝑦


||,

44 Note that equation (9b) is only violated if 100% of endowment can be traded for leisure. But if that is not the case, equation (9c) is violated instead.

Indifference curve Indifference curve

Budget

Budget

c c

𝐿 𝐿

Case A: Convex indifference curve Case B: Concave indifference curve


34

|(𝜂

𝜂

)𝐵

− 1| = ||

(1+𝑦

𝑤𝑀)��𝑐𝑀(1−

𝜕𝑀𝜕𝑤

(−𝜕𝐿𝜕𝑤

))

1−(1+𝑦


||. (A3.18)

APPENDIX 4

I will now derive a point estimator of EMUC for the special case 𝜕𝑃

𝜕𝑤= 0.

Inserting the Slutsky equation (7) into (A3.1) gives:

𝜂 =𝜕𝑀𝑐

𝜕𝑤−

𝜕𝑀

𝜕𝑤𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐

𝜕𝑤

∙(1+

𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

, (A4.1)

𝜂 =𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐

𝜕𝑤−

𝜕𝑀

𝜕𝑤−

𝜕𝐻

𝜕𝑤𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐

𝜕𝑤

∙(1+

𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

, (A4.2)

𝜂 = (1 −𝜕𝑃

𝜕𝑤𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐

𝜕𝑤

) ∙(1+

𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

. (A4.3)

When 𝜕𝑃

𝜕𝑤= 0, (A4.3) reduces to:

𝜂 =(1+

𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

. (A4.4)

(A3.16) gives:

𝜕𝑃

𝜕𝑤< 0

⇒ (1 −

𝜕𝑃

𝜕𝑤𝜕𝐻

𝜕𝑤+

𝜕𝑀𝑐

𝜕𝑤

) > 0, (A4.5)

𝜕𝑃

𝜕𝑤< 0

⇒ 𝜂 >

(1+𝑦

𝑤𝑀)

(1+(1+𝑦

𝑤𝑀)

𝑀

𝐿��𝑐𝐿)

. (A4.6)

Also, because 𝜕𝐻

𝜕𝑤< 0:

𝜕𝑀

𝜕𝑤< 0

⇒

𝜕𝑃

𝜕𝑤< 0. (A4.7)


35

APPENDIX 5

I will now derive an EMUC formula given the following assumption:

𝑢(𝑐, 𝐿) ≈ 𝑓(𝑐) + 𝑘 ∙ 𝐿, (A5.1)

FOC:

𝜕𝑢

𝜕𝑃= (

𝜕𝑐

𝜕𝑀

𝜕𝑀

𝜕𝑃+

𝜕𝑐

𝜕𝐻

𝜕𝐻

𝜕𝑃) + 𝑘

𝜕𝐿

𝜕𝑃= 0,

𝜕𝑢

𝜕𝑃= 𝑓′(𝑐) (𝑤

𝜕𝑀

𝜕𝑃+ 𝑔′(𝐿)

𝜕𝐻

𝜕𝑃) − 𝑘 = 0. (A5.2)

Inserting (A1.3) into (A5.2) gives:

𝜕𝑢

𝜕𝑃= 𝑓′(𝑐) ∙ 𝑤 (

𝜕𝑀

𝜕𝑃+

𝜕𝐻

𝜕𝑃) − 𝑘 = 0,

𝑘 = 𝑤𝑓′(𝑐). (A5.3) Differentiation of (A5.3) w.r.t. 𝑤 gives:

0 = 𝑓′(𝑐) + 𝑤𝑓′′(𝑐) (𝑀 +𝜕𝑐

𝜕𝑀

𝜕𝑀

𝜕𝑤+

𝜕𝑐

𝜕𝐻

𝜕𝐻

𝜕𝑤),

−𝑤𝑓′′(𝑐) (𝑀 +𝜕𝑐

𝜕𝑀

𝜕𝑀

𝜕𝑤+

𝜕𝑐

𝜕𝐻

𝜕𝐻

𝜕𝑤) = 𝑓′(𝑐),

−𝑤𝑓′′(𝑐) (𝑀 + 𝑤 (𝜕𝑀

𝜕𝑤+

𝜕𝐻

𝜕𝑤)) = 𝑓′(𝑐),

−𝑓′′(𝑐)

𝑓′(𝑐)=

1

𝑤(𝑀+𝑤(𝜕𝑀

𝜕𝑤+

𝜕𝐻

𝜕𝑤))

,

𝜂 =𝑐

𝑤(𝑀+𝑤(𝜕𝑀

𝜕𝑤+

𝜕𝐻

𝜕𝑤))

,

𝜂 =(𝑦+𝑤𝑀)

𝑤𝑀∙(1+𝑤

𝑀(

𝜕𝑀

𝜕𝑤+

𝜕𝐻

𝜕𝑤))

,

𝜂 =1+

𝑦

𝑤𝑀

1+𝜀𝑀𝑤(1+

𝜕𝐻𝜕𝑤𝜕𝑀𝜕𝑤

)

. (A5.4)


36

REFERENCES

Aguiar, M., Erik Hurst, E. (2007), “Measuring Trends in Leisure: The Allocation

of Time Over Five Decades”, The Quarterly Journal of Economics 122 (3):

pp. 969-1006, doi: 10.1162/qjec.122.3.969

Anthoff, D., Tol, R. S. J. & Yohe, G. W. (2009), “Risk aversion, time preference, and

the social cost of carbon”, Environmental Research Letters, vol. 4.

Baum, S. D., (2009) “Description, prescription and the choice of discount rates”,

Ecological Economics, vol. 69, issue 1, pp. 197-205, ISSN 0921-8009.

http://dx.doi.org/10.1016/j.ecolecon.2009.08.024.

Blundell, R., & MaCurdy, T. (1999). “Labor supply: A review of alternative

approaches”, Handbook of labor economics, 3, 1559-1695.

Burgoon, B.; Baxandall, P., (2004) “Three Worlds of Working Time: The Partisan

and Welfare Politics of Work Hours in Industrialized Countries”, Politics

Society, 32: 439. DOI: 10.1177/0032329204269983,

http://pas.sagepub.com/content/32/4/439

Chetty, R., (2006), "A New Method of Estimating Risk Aversion." American

Economic Review, 96(5): 1821-1834. DOI: 10.1257/aer.96.5.1821.

http://www.aeaweb.org/articles.php?doi=10.1257/aer.96.5.1821

Evans, D., (2005), “The Elasticity of Marginal Utility of Consumption: Estimates

for 20 OECD Countries”, Fiscal Studies, vol. 26, no. 2, pp. 197–224 0143-

5671

Evans, D., (2006), “The elevated status of the elasticity of marginal utility of

consumption”, Applied Economics Letters, 2004, 11, pp. 443–447.

http://dx.doi.org/10.1080/1350485042000189541

http://dx.doi.org/10.1016/j.ecolecon.2009.08.024

http://pas.sagepub.com/content/32/4/439

http://www.aeaweb.org/articles.php?doi=10.1257/aer.96.5.1821

http://dx.doi.org/10.1080/1350485042000189541


37

Evers, M., De Mooij, R., & Van Vuuren, D. (2008). ”The wage elasticity of labour

supply: a synthesis of empirical estimates”, De Economist, 156(1), pp. 25-

43.

Global Wage Report 2012/13: Wages and equitable growth (2013), Geneva,

International Labour Office, ISBN 978-92-2-126237-4 (PDF)

Huberman, M.; Minns, C. (2007), “The times they are not changin’: Days and

hours of work in Old and New Worlds, 1870–2000”, Explorations in

Economic History, 44, pp. 538–567.

Ihrig, K.S., Moe , J. 2000 “The influence of government policies on informal labor:

implications for long run growth”, De Economist, 148 (3), pp. 331–343.

Morley, J.; Sanoussi, F.; Biletta, I.; Wolf, F. (2010): Comparative analysis of

working time in the European Union., Eurofund, Dublin.

http://www.eurofound.europa.eu/working/studies/tn0803046s/index.

htm

Pawlowskia, T., Breuera, C. (2012): “Expenditure elasticities of the demand for

leisure services”, Applied Economics, vol. 44, issue 26, pp. 3461-3477,

http://www.tandfonline.com/doi/abs/10.1080/00036846.2011.577021

Pencavel, J. (2014): The Productivity of Working Hours, IZA Discussion Papers

No. 8129

Ramey, V.A., Francis N., (2009), ”A Century of Work and Leisure”, American

Economic Journal: Macroeconomics 2009, 1:2, pp. 189–224

http://www.aeaweb.org/articles.php?doi=10.1257/mac.1.2.189

Serres, A., Scarpetta, S. & de la Maisonneuve, C. (2002), Sectoral shifts in Europe

and the United States: how they affect aggregate labour shares and the

http://www.eurofound.europa.eu/working/studies/tn0803046s/index.htm

http://www.eurofound.europa.eu/working/studies/tn0803046s/index.htm

http://www.tandfonline.com/doi/abs/10.1080/00036846.2011.577021

http://www.aeaweb.org/articles.php?doi=10.1257/mac.1.2.189


38

properties of wage equations, OECD Economics department working

papers NO. 326, http://www.oecd.org/eco

Statistics Sweden, Table: ”Inkomststruktur för hushåll med personer 20-64 år,

medelvärde, tkr efter inkomstkomponent, sysselsättning och år”,

http://www.statistikdatabasen.scb.se/pxweb/sv/ssd/START__HE__HE0

103__HE0103A/InkStrukt46/table/tableViewLayout1/?rxid=48d6bb10-

5f13-41de-9201-948fab2ea58e (retrieved 2015-09-29)

Thompson, C.S.; A.W. Tinsley, (1979), “Income Expenditure Elasticities for

Recreation: Their Estimation and Relation to Demand for Recreation”,

Journal of Leisure Research, 10 (4): pp. 265–270.

Vandenbroucke, G., (2009), “Trends in Hours: The U.S. from 1900 to 1950”,

Journal of Economic Dynamics and Control, vol. 33, no.1, pp. 237-249.

Weagley, R. O.; Huh, E., (2004), “The Impact of Retirement on Household Leisure

Expenditures”, Journal of Consumer Affairs, 38: 262–281. doi:

10.1111/j.1745-6606.2004.tb00868.x

U.S. Census Bureau, Current Population Survey, Annual Social and Economic

Supplements. Tables P36 and P38. http://www.census.gov/ (retrieved

2015-09-28)

http://www.oecd.org/eco

http://www.statistikdatabasen.scb.se/pxweb/sv/ssd/START__HE__HE0103__HE0103A/InkStrukt46/table/tableViewLayout1/?rxid=48d6bb10-5f13-41de-9201-948fab2ea58e



http://www.census.gov/

Documents

Household Production and the Elasticity of Marginal ...vti.diva-portal.org/smash/get/diva2:952369/FULLTEXT01.pdf · Household Production and the Elasticity of Marginal Utility of