1 Introduction

This paper tests the effect that income uncertainty has on consumption of Spanish households using panel data to obtain a proxy of income uncertainty. I use the Spanish Survey of Household Finances (EFF hereafter), an official survey provided by the Bank of Spain which provides information about different aspects of the economic and financial situation of Spanish households for several years.

This work analyzes whether uncertainty affects household consumption in 2014, the considered as the end of the Great Recession in our country. The Spanish Great Recession is the economic crisis that began in 2008 and ended in 2014, according to the national accounts, based on data elaborated by the National Institute of Statistics (INE). Unemployment, which was at an all-time low in the spring of 2007 (7.95%), reached an historic high in the first quarter of 2013 (27.16%). The GDP growth was negative since the beginning of 2009 until the first quarter of 2014 when started to register positive values again. Wages and salaries also registered a negative growth rate during the considered period of crisis recovering a positive growth path at the beginning of 2014. According to the data of the Bank of Spain, public debt (as % of GDP) suffered an increase of 65 percentage points between 2008 and 2014 when started to recover previous values. The evolution of these indicators, among others, allow to establish the period of the Spanish Great Recession between 2008 and 2014, even when it took more years for the Spanish economy to recover to pre-crisis levels for some of the macroeconomic variables.

The Great Recession was the most important shock on our economy over the last years. During that period uncertainty borne by households increased and for that reason I have decided to analyze the effect it has on household consumption at the end of the crisis. To do that I exploit the panel component of the survey for deriving a measure of income uncertainty (using the individual data on income for a period of 8 years, 2007–2014) and then test whether uncertainty affected household consumption in 2014.

The existence and strength of the precautionary motive for saving, as well as which is the most appropriate measure to proxy the uncertainty, is an unresolved question in the empirical literature testing the precautionary saving hypothesis (for a survey about precautionary saving, see Browning and Lusardi 1996; Attanasio and Weber 2010; Lugilde et al. 2018; or Baiardi et al. 2020). Among the three approaches used in empirical works to estimate the importance of precautionary saving: reduced form estimates, simulation models and subjective expectations, this paper follows the first one and uses objective data to estimate income uncertainty. In particular, the paper is framed in the empirical works which proxy income uncertainty using observed life-cycle income variation and the variability of income (Kazarosian 1997; Carroll and Samwick 1998; Guariglia 2001; Ventura and Eisenhauer 2006).

Although most empirical works find evidence of an effect of uncertainty on savings, not in many countries there is evidence enough about this motive for saving (the US, Italy, the UK, Germany and few others). In the case of Spain, there is not too much evidence about precautionary saving and most of the papers proxy the uncertainty trough unemployment risk (Albarrán 2000; Barceló and Villanueva 2010; Barceló y Villanueva 2016; Campos and Reggio 2015; Lugilde et al. 2018).

The main feature of this paper is to provide evidence about precautionary saving in Spain exploiting the panel component of the survey for deriving a measure of income uncertainty. The analysis is performed in two steps. In a first step I estimate a measure of income uncertainty based on panel data from 2007 to 2014. In particular, I calculate the average household real income over the period and its standard deviation for each household as proxies of household normal income and income uncertainty, respectively. Related to this, I show that this measure correlates with some variables that are commonly thought to be related to uncertainty, like self-employment, age, etc. In a second step, I relate the variable of income uncertainty to consumption, testing whether uncertainty affects household consumption in 2014.

Since the econometric results show a negative impact of uncertainty on household consumption, I can conclude about the evidence of the existence of precautionary saving in Spain. The results obtained weakly differ depending on the consumption variable used as dependent variable in the model. When using the logarithm of the household consumption, I obtain that a 1% increase in the income uncertainty will decrease consumption in about 8.8%. However, when I use the ratio between consumption and average income the effect is lower, given the average normal income and consumption in the sample, consumption will decrease by 8.1%. To the best of my knowledge, it is the first paper concluding about the elasticity of consumption regarding income uncertainty. Other works test the impact of income uncertainty on household consumption (see Carroll 1994; or Miles 1997), but their results are not in elasticity terms.

This paper provides empirical evidence about precautionary savings in Spain after a convulsed period such as the Great Recession years in which households had to adjust their consumption patterns. Given the little evidence of that for this country this could be considered the main contribution of this paper. Moreover, to the best of my knowledge, this is the first paper providing evidence about precautionary saving in Spain measuring income uncertainty from observed household real income data during a period.

After this introduction, the paper is organized as follows. Section 2 briefly summarizes the existing literature about precautionary saving and the available empirical evidence for Spain. Section 3 provides a description of the constructed uncertainty measure and shows that this measure correlates with some variables that are commonly thought to be related to uncertainty. Section 4 presents the econometric model and the results. Finally, Section 5 concludes.

2 Review of the literature

When consumption decisions are made under uncertainty, and individuals are prudent and seek protection from risk, there is a significant negative impact on current consumption. Under some specific properties of the utility function (utility is increasing and concave and marginal utility is convex) uncertainty generates a positive extra-saving, the so-called “precautionary saving” (Leland 1968).

The relevance of this motive for saving is an issue addressed mainly empirically. Despite the large literature on that, there is no consensus about the intensity of the precautionary reason for saving, nor on which is the most appropriate measure to approximate the uncertainty.

Three approaches to estimate the importance of precautionary saving have been used in empirical works: reduced form estimates, simulation models and subjective expectations. Works following the first approach attempt to estimate the impact of income uncertainty on the reduced forms of consumption or wealth; that is, to estimate reduced form equations inspired by the PIH model with a role for precautionary saving (Miles 1997; Kazarosian 1997; Lugilde et al. 2018; among others). This approach also provides evidence in favour or against precautionary saving but does not deliver estimates of the parameters of the utility function (such as the coefficient of relative prudence).

Studies following the second approach estimate the paths of consumption and wealth in models with precautionary saving, matching simulated data to observed wealth and consumption distributions. Structural estimations deliver estimates of the parameters of the utility function but require the utility function, the budget constraint, the sources of risks, and the income process to be specified. Pioneering in this approach are Gourinchas and Parker (2002) and Cagetti (2003), who calibrate an explicit life cycle optimization problem by using empirical data on the magnitude of household-level income shocks, and search econometrically for the values of parameters such as the coefficient of relative risk aversion that maximized the model’s ability to fit some measured features of the empirical data. The intensity of the precautionary motive emerges, in each case, as an estimate of the coefficient of relative risk aversion.

Another direct strategy to analyse the existence of precautionary saving is the use of subjective expectations from survey questions data (Lusardi 1997; Guiso et al. 1992; Mastrogiacomo and Alessie 2014; or Christelis et al. 2020). The literature based on subjective expectations attempts to avoid the problem of lack of information that is not observed by the econometrician by asking people to report quantitative information on their expectations. This literature relies on survey questions to elicit information on the conditional distribution of future income, and measures shocks as deviations of actual realizations from elicited expectations. Hayashi’s (1985) is the first study to adopt this approach. Another use to subjective expectations is to measure expected consumption growth and expected consumption risk in Euler equation with precautionary saving using survey data that record respondents’ own assessments of these variables. This is an alternative method to the simulation models to directly test precautionary saving through the estimation of the relative prudence coefficient.

This paper is framed within the empirical works which proxy income uncertainty using observed life-cycle income variation (Kazarosian 1997; Carroll and Samwick 1998; Guariglia 2001; Ventura and Eisenhauer 2006). In particular, the uncertainty is proxied through the income variability following the first mentioned approach (reduced form estimates) and using objective data to estimate income uncertainty. Since the main prediction of the precautionary saving model, with respect to the life cycle–permanent income model, is that saving and wealth are related not only to the first moment, but also to higher moments of income, a wide branch of the literature has estimated uncertainty by the income variability. Consumers accumulate not only to offset future declines in income, but also to insure against income uncertainty, proxied traditionally by the standard deviation or the variance of income. There have been several methods to deal with the measurement of income uncertainty in the works using objective micro data.

A popular method with cross-section data is to use an aggregate estimate of income variance by categorizing sample observations into groups according to socio-economic characteristics, e.g., occupation, age, education, etc. (Dardanoni 1991). The within-group income variance is then used as a proxy for individual income variance (Dardanoni 1991; Miles 1997; or Mishra et al. 2013; follow this approach). To be valid, this method requires each individual to rely on the same set of variables to form expectations and the individuals and the econometrician to have the same information set.

Some works using panel data use not only the information from panel but also the individual characteristics in order to derive a measure of income uncertainty (see Carroll 1994; Kazarosian 1997; or Guariglia and Rossi 2002). The use of panel microdata allows to directly test whether individuals change their behaviour when there are changes in the uncertainty they bear according to theoretical predictions. For that, other works exploit the panel structure of the data to calculate the permanent/normal income from the household real income over a period and the variance of this income (see, for example, Carroll and Samwick 1998; Guariglia 2001; or Ventura and Eisenhauer 2006). Carroll and Samwick (1998) include the log of the variance of the log of income as an atheoretical measure of uncertainty (besides the log of relative Equivalent Precautionary Premium) and find that both coefficients are highly significant for all three measures of wealth considered (very liquid assets; non-housing non-business wealth and total net worth). Guariglia (2001) uses British Household Panel Survey (BHPS) data (years 1991–1998) to estimate three household specific measures of earnings uncertainty and test precautionary saving.Footnote 1 The first of them is obtained taking the square of the difference between detrended household earnings in 1991 and 1998 and dividing it by seven to have an annual rate. The second is the variance of income, \({Y}_{t}\), over the eight available waves (this measure assumes that all income shocks are transitory). The last measure is the variance of income over waves two to eight (variance of \({Y}_{t}-{Y}_{t-1}\) and contrary to previous assumes that all income shocks are entirely permanents). Guariglia concludes that there is a strong precautionary motive for saving for all measures of uncertainty employed. Ventura and Eisenhauer (2006) use the Survey of Household Income and Wealth (SHIW) to analyze three principal saving motives: intertemporal saving, bequest motive and precautionary saving.Footnote 2 They select households with income reported in both 1993 and 1995, and among them they focus only on savers. To capture the precautionary motive, for each household, they calculate the average real income and its variance between these two years, which they use initially as proxies of current income and uncertainty, respectively, in a saving equation. Then, exploiting the estimated regression coefficients as well as mean values of the variables, they calculate point estimates of absolute and relative prudence, and obtain that each young household saved 15.2% of its total annual saving by precautionary purposes.Footnote 3

In this paper I also make use of the panel component of the survey and perform the analysis in two stages. Firstly, by exploiting the panel structure, I calculate the average household real income over the period 2007–2014 and its standard deviation for each household. Then I use these variables as proxies of household normal income and income uncertainty, respectively, in a cross-sectional regression of consumption (Guariglia 2001; follows a similar strategy). The underlying assumption is that individuals use their own past incomes to forecast their future income and have rational expectations. As pointed by Dynan (1993), the household consumption changes only in response to unexpected changes in income (Dynan 1993, pag. 1105) so, in this paper I test the existence of precautionary saving by analyzing the effect of the uncertainty on consumption (see Dardanoni 1991; Dynan; 1993; Carroll 1994; Miles 1997; Banks et al. 2001; Benito 2006; or Lugilde et al. 2018; among others).

Although most empirical works find evidence of an effect of uncertainty on savings, most of them are focused on the same economies, so there are not many countries where there is evidence about precautionary saving (the US, Italy, the UK, Germany and few others). In the case of Spain, there is a little evidence about the existence of precautionary saving. Albarrán (2000) uses micro-data from a rotating panel, the Spanish Family Expenditure Survey, to analyze precautionary saving associated with income uncertainty. He finds that consumption growth is not affected by household-specific risk but by cohort-specific and aggregate risk. Barceló and Villanueva (2010) using data from the EFF find evidence in favour to the existence of precautionary savings proxying the probability of losing employment by the type of contract that the main recipient of income at the household has. In a following paper, Barceló y Villanueva (2016), using the same survey, analyze the effect the changes in severance payment have on wealth accumulation and find that older workers covered by fixed-term contracts accumulate more financial wealth than other workers. Campos and Reggio (2015), using consumption panel data, find that households reduce consumption in response to the realization of negative news on future income growth contained in the unemployment rate (calculated from the Spanish Labour Force Survey according to the level of education and age of the primary earner in the household). Lugilde et al. (2018) use the Spanish EFF and the Labour Force Survey and find that subjective measures generate a non-significant impact on consumption, and hence on saving, and the impact the objective measures have is different depending on the moment of the business cycle. Only in a context when unemployment is high and rising, it becomes an important source of uncertainty while the job insecurity that the household reference person faces generates a significant negative impact on consumption at all business cycle horizons as well as regardless of the econometric specification. Therefore, to provide empirical evidence about the effect the uncertainty has on consumption for the Spanish households is an important contribution of this paper.

3 Measuring income uncertainty from the EFF data

In the context of precautionary motive for saving the use of microeconomic panel data is preferred to analyze consumption behaviour since it allows to capture the effects of individual income uncertainty along a specific period. For this reason, to perform the analysis of precautionary saving in Spain I use the EFF data. It is an official survey compiled by the Bank of Spain, which has been published since 2002 (every three years) to obtain direct information about the financial conditions of the Spanish families. It is the only statistical source in Spain that allows the linking of incomes, assets, debts and expenditure of each household.Footnote 4 The survey of Banca d’Italia, Survey on Household Income and Wealth (SHIW), and the Survey of Consumer Finances (SCF) of the US Federal Reserve were the models that inspired this survey.

This paper focuses on the panel component of the survey to analyze the existence and strength of precautionary saving in Spain. Since I want to consider the normal income of the household, I work with a balanced panel including the households participating in the survey since 2008 for which eight years of income information is available.Footnote 5 The balanced panel comprises 1524 Spanish households.

The variable of household income is provided in the survey data and it is constructed aggregating the data of individual income of household members, the income obtained from assets and the non-labour income received by the whole household. Therefore, the income variable is the total gross income of the household, i.e. before taxes and social-security contributions. The income data are available for years 2007, 2008, 2010, 2011, 2013 and 2014 and expressed in real terms (2014 euros) using the Consumer Price Index (CPI) as deflator.Footnote 6 From this information, exploiting the panel component, I calculate the household average income over the whole period (2007–2014) and from that the standard deviation of the household income (Guariglia (2001) follows a similar approach to calculate the income uncertainty measures). These variables are used as proxies of the household normal income \((\overline{Y })\) and income uncertainty \((SDY)\), respectively. The use of an 8-year period to estimate the household normal income is because of the focus of this paper. Since the aim is to analyze the consumption/saving behaviour of Spanish households at the end of the Great Recession, using income information covering the crisis period seems to be the best way to address it.

From the household average income, I construct a control variable capturing if the household income was under a threshold defined as 20% of the average income of the period in some years (Deidda (2013) establish the same income threshold and excludes from the analysis the households whose earnings were under it). Only 4.54% of the households in the sample had income under 20% of its average income in some of the previous years and only for 1.73% of the households the current income, income of 2014, is under the threshold. I include this variable in the consumption regressions to check if that has some effect on consumption and if it varies depending on the moment in which that occurs, 2014 or some of the previous years.

The estimated measure of income uncertainty \((SDY)\) captures current income shocks but, under the assumption that individuals use their own past incomes to forecast their future income and have rational expectations, it is a good proxy for future uncertainty. The supposition is that those households with high income variability could have also more uncertainty about future income as a projection of this income uncertainty that they have. In fact, the estimated measure of income uncertainty \((SDY)\) correlates with some variables commonly related to uncertainty. To show that, I include below several graphs collecting the relationship between the standard deviation of income, \(SDY\), and different characteristics of the household reference person commonly related to uncertainty. In this survey the reference person is self-determined and can be defined as the person, or one of the persons, responsible for the accommodation (it will normally be the person in the household who chiefly deals with the financial issues) (Fig. 1).

Fig. 1
figure 1

Relation between the SDY and the age of the reference person. Own elaboration from the EFF balanced panel data

Full size image

The youngest and the elderly exhibit a higher standard deviation of income in relation to the average income, this being more acute for the former (especially for those between 26 and 35 years old). This is consistent with the idea that households with younger heads respond more strongly to the income uncertainty because households with the youngest household heads need to save more in order to build a buffer stock of savings and retired individuals face uncertainty with respect to their survival as well as medical and nursing home expenses which are not present in the middle-aged (Chamon et al. 2013; Kopecky and Koreshkova 2014) (Fig. 2).

Fig. 2
figure 2

Relation between the SDY and the education of the reference person. Own elaboration from the EFF balanced panel data

Full size image

Among the different levels of education those with a “post-graduate” have the highest standard deviation of income (being, in average, 54% of their average normal income) while those with “primary education” have the lowest. This could be related to the labour status or the occupation that individuals are performing in function of their educational attainment (Fig. 3).

Fig. 3
figure 3

Relation between the SDY and the labour status of the reference person. Own elaboration from the EFF balanced panel data. Notes: “other inactivity” refers to “Permanently disabled or unable to work”, or “Student” or “Housewife/house husband”

Full size image

As expected, those households whose reference person is “self-employed” jointly with those in which it is “permanently disabled or unable to work, or student or housewife/house husband” (that is, “other inactive”) have the highest uncertainty while the “employees” have the lowest. This is in line with the assumed in the literature: the self-employed have presumably higher income uncertainty (Leland 1968; Deidda 2013; Mishra et al. 2013) (Fig. 4).

Fig. 4
figure 4

Relation between the SDY and the occupation of the reference person. Own elaboration from the EFF balanced panel data. Notes: Occupation is displayed for those who are working (568). Managers: “in the public or private sector”. Service: “Hotel and catering, personal, security and sales services”. Skilled craftsman or worker in manufacturing, construction, or mining industries. Plant & machine operators: “and assemblers”

Full size image

Among the different occupations, “managers in the public or private sector”, followed by those employed in the category “skilled workers in agriculture and fishing”, have the highest income uncertainty in respect to their average normal income (for them the mean of the SDY is more than 50% of the mean of the normal income). In the sample considered, these occupations are also those with the highest ratio of self-employed and the managers are people with highest educational level. “Skilled craftsman or worker in manufacturing, construction or mining industries” jointly with “plant and machine operators and assemblers” have the lowest income uncertainty and have also a lower level of education than managers (nobody in these occupations is a postgraduate) (Fig. 5).

Fig. 5
figure 5

Relation between the SDY and the defined threshold for household income. Own elaboration from the EFF balanced panel data

Full size image

As regards the variable capturing whether household income was under 20% of normal income in some years, the graph shows that those whose income in 2014 was under the threshold defined have the highest uncertainty in respect to their normal household income, while those whose income was always over the threshold have the lowest. That supports the adequacy of our proxy of uncertainty for capturing the uncertainty effect on household consumption.

The graphs show the expected relation between the standard deviation of the income (in relation to the normal income) and the different variables, supporting thus the use of this variable as proxy for the income uncertainty borne for the household.

Once the validity of the estimated uncertainty proxy is shown, the following section tests the effect that the uncertainty about future income, measured trough the standard deviation of household income, has on household total consumption in the year 2014.Footnote 7

4 Econometric model and results

The econometric model relates the consumption of a household with several covariates related to the personal, family, work and financial characteristics of the households included in the sample. Specifically, assuming that the relationship among the dependent and independent variables can be expressed in a log-linear form, the models are:

$$log C_{i,t} = \beta_{0} + \beta_{1} log\overline{Y}_{i} + \beta_{2} logSDY_{i} + \gamma Z_{i,t} + v_{i,t} .$$
(1)
$$C_{i, t} /\overline{Y}_{i} = \alpha_{0} + \alpha_{1} log\overline{Y}_{i} + \alpha_{2} logSDY_{i} + \theta Z_{i,t} + \varepsilon_{i,t}$$
(2)

where \({\beta }_{0}\), \({\alpha }_{0}\) are the intercept; \(\gamma\), \(\theta\) are vectors of parameters to be estimated; \({Z}_{i,t}\) is a vector of variables that reflect the main individual characteristics of each individual/household and the main economic determinants of consumption at time \(t\) (income, real and financial wealth, debt, risk aversion, family composition, age and education level of the reference person); \({\overline{Y} }_{i}\) is the household average income over the period (2007–2014); \({SDY}_{i}\) is the standard deviation of household income (the proxy of uncertainty) and \({v}_{i,t}\), \({\varepsilon }_{i,t}\) are the error terms assumed independently and identically distributed as a \(N\left(0, {\sigma }^{2}\right)\). The regressions are estimated for the last year, 2014, to analyze how the average income of the period and the standard deviation of the income affect the household consumption in this year (therefore, \(t=2014\)). The economic variables are expressed in logarithms and refer to the whole household.Footnote 8 The age and the educational level are those of the household reference person. The difference between both models is the dependent variable, in \((1)\) \({logC}_{i,t}\) is the logarithm of consumption of the i-th household in 2014 while in \((2)\) \({C}_{i,t}/{\overline{Y} }_{i}\) is the ratio between consumption of the i-th household in 2014 and the average income of the i-th household for the period 2007–2014. The equations are estimated by OLS.

Therefore, I assess the existence of precautionary saving by analyzing the effect of household income uncertainty on consumption. If there is a precautionary saving, uncertainty in the current period (approximated by the standard deviation of income, \({SDY}_{i}\)) should increase savings and thus decrease current consumption, i.e., a negative sign on this uncertainty variable is expected.

Table 1 shows the results of the estimations for 2014. Columns (2) and (4) summarize the estimation of the two consumption models including the uncertainty measure. In particular, column (2) shows the results for the model using the logarithm of total consumption as a dependent variable while column (4) summarizes the results for the model using the ratio between consumption and the average income as a dependent variable. Columns (1) and (3) summarize the estimation of both consumption equations without any uncertainty measure to provide a baseline model. In general, the variables included in the estimations are significant (and show the expected signs) and the regressions have a relatively high goodness of fit, with an \({R}^{2}\) around 62% in the logarithm of consumption equation and about 34% for the equation of consumption-average income ratio, and the F-statistic suggests that the null hypothesis of jointly insignificance should be rejected.

Table 1 The uncertainty effect on household consumption in 2014
Full size table

In general, the results for the standard control variables are in line with previous analyses, with expected signs. Wealth impacts positively on consumption, both real and financial wealth. During the period considered households tended to accumulate financial assets. According to the Bank of Spain, as compared to the first quarter of 2009, in 2014 the percentage of Spanish households with some financial assets was greater although the decrease in this percentage from 2011 (the increase in this percentage between 2009 and 2011 was higher in the lower half of the wealth distribution but also its decrease between 2011 and 2014 was greater for this group). For families with some kind of financial asset, the median value of these assets increased by 23.1% between 2009 and 2011 but decreased by 5.1% between 2011 and 2014.Footnote 9

Income is significant in both specifications and the elasticity of income remains more or less stable between the baseline specification of the model and the specification with uncertainty, which means that the estimated parameter is robust to the type of specification. But the sign of the variable change with the dependent variable: in the model with the logarithm of consumption as a dependent variable, the income has a positive effect on consumption while in the model of the ratio between consumption and average income the impact is negative. It shows that as income increases, the propensity to save increases (or MPC declines). Since the magnitudes of the coefficients in the model for the logarithm of consumption are lower than 1, therefore, as income increases, consumption goes one, but elasticity is less than one.

The dummy variable reflecting whether the household is risk averse has a negative and significant coefficient in both models. However, the dummy collecting the existence of credit constraints in the households is not significant in any of the specifications. The relation between the level of indebtedness and household consumption is positive (values close to zero) but not significant.

Households whose income was under the threshold (defined as 20% of the average income of the period) in some years before 2014 had a higher consumption that year (but with no significant coefficients), while those whose income was under the threshold in 2014 reduced their consumption (with significant coefficients). Household characteristics show the relations expected. Additionally, the estimated coefficients are, in general, robust to the specification as regards the inclusion of the uncertainty measure, even though they differ in magnitude between the two consumption models considered in the analysis.

In relation to the uncertainty measure, the standard deviation of household income shows a negative and significant coefficient in both models. Therefore, an increase in the income uncertainty borne by the households reduces its current consumption, implying (given the level of household income) certain amount of precautionary savings. This result is in line with those of Albarrán (2000), Barceló and Villanueva (2016), Campos and Reggio (2015) or Lugilde et al. (2018) who also show evidence of precautionary saving in Spain in different periods of time and using different data sources. The main difference with these works is that I use an uncertainty measure derived from observed household income from panel data and most of the evidence about precautionary saving for Spain estimate unemployment risk or use rotating panel data. The effect the income uncertainty has on consumption is softer when taking into account the level of income than in absolute terms. The uncertainty measure has a larger impact on the logarithm of consumption (− 0.088) than on the ratio consumption—normal income (− 0.026). This reduction of 0.026 in the ratio \(Cons /\overline{Y }\) when the \(SDY\) increases by 1% implies, given the average consumption and normal income in the sample, that consumption will decrease by 8.1% while in the model for the \(ln\left(Cons\right)\), an increase of 1% in the \(SDY\) will decrease consumption by 8.8%.Footnote 10

These results go in line with other works showing a strong precautionary motive for saving. Thus, for the UK, Dardanoni (1991) claims that more than 60 percent of savings arise as a precaution against future income risk, Miles (1997) concludes that doubling income variability for the typical household reduces consumption by, on average, almost 5% or Guariglia and Rossi (2003) also a strong precautionary motive for saving. Using US data, Carroll’s (1994) results suggest that a one-standard-deviation increase in uncertainty decreases consumption from 3 to 5 percent; Carroll and Samwick (1998) find that between 32 and 50% of wealth in the sample is attributable to the extra uncertainty or Kazarosian (1997) obtains that a doubling of uncertainty increases the ratio of wealth to permanent income by 29%. Ventura and Eisenhauer (2006) obtain that on average, each young Italian household saved 15.2 percent of its total annual saving, by precautionary purposes while Lusardi (1997) finds that precautionary accumulation ranges from 16 to 24% of total wealth accumulation.

In the case of Spain, the scant evidence shows mixed results. Albarrán (2000) concludes that a 1% increase in the variance of cohort component of risk would imply that households raise their precautionary saving by 2.8% and Campos and Reggio (2015) find that a one-point increase in the unemployment rate was related to a reduction in household consumption of more than 0.7% per equivalent adult. However, Barceló and Villanueva (2016) obtain a stronger effect and conclude that mature temporary workers save an extra 40% of income in liquid assets and Lugilde et al.’s (2018) results show that the labour conditions that the individuals face in the workplace may become an important source of uncertainty.

To the best of my knowledge, it is the first paper concluding about the elasticity of consumption with respect to income uncertainty. The results show that an increase of 1% in the standard deviation of household income decreases household consumption between 8 and 8.8%. From my point of view, this large reaction of consumption can be explained in part because of the macroeconomic context (end of a crisis -after a period of expansion and economic growth- which made families worry about uncertainty) and is in line with works finding and strong precautionary motive for saving.

5 Concluding remarks

Earnings uncertainty is the source of uncertainty most frequently studied in the theoretical literature about precautionary savings and income variability is the most common uncertainty proxy used in empirical works. This paper contributes to the existing literature by testing the effect that income uncertainty has on the consumption of Spanish households. The main contribution of this paper is to provide evidence about precautionary saving in Spain by measuring income uncertainty from the panel component of the EFF. I derive a measure of income uncertainty by using the individual observed data on income for a period comprising 8 years. From that I calculate the standard deviation of household real income as proxy of the income uncertainty borne by the household and test the effect that it had on household consumption in 2014.

According to the estimations, an increase of 1% in the standard deviation of household income (uncertainty measure) reduced its current consumption between 8.1 and 8.8% implying (given the level of household income) a certain amount of precautionary saving. Therefore, I can conclude about the evidence of the existence of precautionary saving in Spain. This evidence for the Spanish households is consistent with the hypothesis that households adjust their consumption and savings to changes in the uncertainty of income.

In addition, to the best of my knowledge, this is the first paper showing evidence about precautionary saving in Spain measuring income uncertainty from observed household real income data during a period of time (Carroll 1994; or Miles 1997; also test the effect the income uncertainty on consumption but for US and UK) and also concluding about the elasticity of consumption in respect to income uncertainty.