Household sentiment and the stock holding puzzle
Gelein Cornelis Huiskes
Master thesis Msc. Finance and Msc. Economics
University of Groningen
Supervisor: prof. dr. R.J.M. Alessie
June 2015
Abstract
By utilizing the Dutch Household Survey with around 1400 households, this paper finds evi-dence of a positive relation between sentiment and stock ownership on a micro level. We test both a binary dependent variable model for holding risky assets and level dependent variable model for the level of riskiness of financial wealth. Household sentiment levels are constructed in a similar fashion as the Michigan Consumer Sentiment Index (ICS).
JEL-Classification: D12, D14,
Keywords: Household behaviour, Stock holding puzzle, Risk aversion, Dutch Households
1
Introduction
Half a century ago households viewed the aspects of the economy through the single dimension of how it effected their job and income prospects. Currently, it is expected that households view their economic prospects in a broader view. This includes expected inflation, expected returns of financial assets, real estate, pension and health entitlements (Curtin, 2007). According to standard economic theory and historical performance of equity markets, almost all households should at least keep a small portion of their wealth in stocks or other risky assets. However, data shows that this is not the case at all. In the United states, among the highest concerning household stock ownership, 56% percent of households have money invested in the stock markets, in Germany and the United-Kingdom respectively 6.3% and 23% in 20121. This lack of participation of households in equity markets is coined the ”stock holding puzzle” by Haliassos and Bertaut (1995). For its importance in analyzing life-cycle behaviour, especially retirement planning, understanding household stock ownership is important.
Much effort has been put in solving this puzzle. Guiso, Haliassos, and Jappelli (2003) show that stock holdings strongly depends on wealth and age. Extending the standard model with oa. transaction and holding cost (King & Leape, 1998), background risk (Kimball, 1993) and labour supply flexibility (Bodie, Merton, & Samuelson, 1992) has contributed answering this puzzle. Other papers find evidence of a link between financial literacy and the lack of equity positions among households. Households shy away from away from financial markets because they have little knowledge of stocks and stock markets (van Rooij, Lusardi, & Alessie, 2011).
Recent research has started to elicit private households expectations of stock market returns (Dominitz & Manski, 2007). Households beliefs about future events play a central role in
looking models of decision making. Hurd, van Rooij, and Winter (2011) report findings that the expectations of the stock market are correlated with stock ownership.
This paper investigates the relationship between stock market participation by households and sentiment. We seek to find evidence of participation in financial markets driven by household sentiment. For decades there has been research on the role that consumer sentiment indices (CSI)2, have on aggregate consumption. Initially, research whether CSI’s have an autonomous role in forecasting led to mixed results (Golinelli & Parigi, 2004). Some show that indeed the indices play an autonomous role (see Mueller, 1963; Adams, 1964), other research shows that these indices could be seen as nothing more than a synthesis of macroeconomic indicators (see Hymans, Ackley, & Juster, 1970; Shapiro, 1972). The issue of the role of CSI is still discussed, although the prevailing opinion now seems to be that it helps to predict the evolution of economic activity (Golinelli & Parigi, 2004). This papers trans-locates this forecasting power of consumer sentiment to the stock owner puzzle. Where we expect consumer/household sentiment to have a positive effect on the ownership of risky assets.
We use data of about 1350 Dutch households from 2002 till 2014. Using the Dutch Household Survey (DHS) we are able to collect psychological characteristics and data on financial wealth at the household level. By constructing a sentiment index at the household level we are able to test the relation between household sentiment and individual financial behavioural. This paper adds to the literature by assessing the effect of consumer sentiment on stock market participation at the household level after accounting for macro economic trends. Whereas, the majority of the literature looks at financial behaviour and decision making at an aggregated macroeconomic level.
overview will be given of existing literature and the economic theory behind this research. Followed by, a description of the used data in section 3. After which we construct our testable econometric model in section 4. In section 5 we will present the results. Lastly we will provide a conclusion in part 6.
2
Economic theory and existing literature
This part of the paper will give an overview of relevant literature concerning the relationship between household sentiment and financial behaviour. Also will the economic methodology behind the model be described, and the hypotheses to be tested will be presented.
and institutional investors. Households seem to irrationally buy attention grabbing stocks. However, Jansen and Nahuis (2003) find a strong positive correlation between stock return and changes in sentiment of a group of Europeans countries, but they find that higher returns Granger-cause sentiment on very short-term horizon. Jansen and Nahuis (2003) find that stock market-confidence relationship is driven by expectations about economy-wide conditions rather than personal finances. A finding also put forward by Otoo (1999). Furthermore, she finds that the increase in sentiment, due to a rise in stock prices, can be contributed to the fact that people see equity prices as a leading indicator of future economic activity. Not because of personal wealth effects.
The relationship between investing and sentiment is also investigated at a micro level. Amromin and Sharpe (2009) suggest that for most retail investors, forward-looking Sharpe ratios are unequiv-ocally higher when the economy is expected to be strong, sentiment thus plays a role in financial decision making by households. This reduction in perceived risk can lead to more household enter-ing financial markets and buyenter-ing risky assets. There is a consistent significant relation between self attributed risk attitude and stock ownership by households (Shum & Faig, 2006).
Manzan (2007) argue that there are at least three different models for stock market expectations. Individual might believe that prices follow a random walk with drift; believe in means reversion of stock prices; or they might believe in persisting recent price changes.
Returning to the stock holding puzzle. Using data of Dutch households Hurd et al. (2011) suggest that the low stock market participation of these households is due to a combination of perceived risk and pessimism. We follow this finding in constructing our first hypothesis. Where higher sentiment leads to a reduction in pessimism(optimism) and thus a higher stock market participation rate.
Hypothesis 1 :There is a positive significant relationship between household sentiment and stock ownership.
This means that households with relative high sentiment will be more likely to hold risky assets. In other words, the lower perceived risk due to higher sentiment leads to higher participation levels in equity markets by households. The second hypothesis we state is the following:
Hypothesis 2 :Given the household owns risky assets, there is a positive significant relation be-tween household sentiment and the ratio of risky versus total financial wealth.
By testing this hypothesis we aim to prove that higher household sentiment leads to riskier asset portfolios among the households that already were involved in equity markets. Riskier asset portfolios in the form of a higher ratio of risky assets versus total financial wealth.
3
Data
In this paper use is made of data of the DNB (Dutch national bank) Household Survey. The survey started in 1993, but this paper uses data starting in the year 2002 because of consistency problems in the survey. In order to create the sentiment index we used part 7 of the survey: the questionnaire on economic and psychological concepts. The questions used are similar to those asked for the Index for Consumer Sentiment (ICS)of Michigan Survey Research Center (2015) by the University of Michigan (see appendix A1). Below a description of this index and the one used in this paper is presented.
3.1
Sentiment Index
Almost all sentiment indexes, are constructed in the following manner. They allow for 5 possible answers to their questions. Ranging from very positive, positive, neutral, negative to very negative. Two steps are involved in the computation of the overall sentiment index (Curtin, 2007). Fist, for each question a balance score is computed based of the following percentage distribution. Let PP be the proportion of very positive responses, P the proportion of positive responses, M the proportion of negative responses and MM the proportion of very negative responses.
B = (P P +1 2P ) − (
1
2M + M M ) (1)
The overall consumer sentiment index is the average of the balance scores:
CSI3= n
X
i=1
Bi/n (2)
Where n denotes the number of questions and i the question.
Since the DHS does not provide the exact same questions, the following question from the psychological section of the DHS dataset are used:
• fX1= INKNORM = ”Is this income unusually high or low compared to the income you would
expect in a regular year, or is it regular?” Possible answers: 1) unusually low, 2) regular, 3) unusually high)
• fX2 = INKROND = ”How well can you manage on the total income of your household?”
Possible answers: 1) it is very hard, 2) it is hard, 3) it is neither hard nor easy, 4) it is easy, 5) it is very easy)
• fX3= FINSITU = ”How is the financial situation of your household at the moment?” Possible
answers: 1) there are debts, 2) need to draw upon savings, 3) it is just about manageable, 4) some money is saved, 5) a lot of money can be saved)
• fX4 = INKEVEN = ”Over the past 12 months, would you say the expenditures of your
household were higher than the income of the household, about equal to the income of the household, or lower than the income of the household?” Possible answers: 1) the expenditures were higher than the income, 2) the expenditures were about equal to the income, 3) the expenditures were lower than the income)
• fX5 = INK25A = ”When you think of the next 12 months, do you think the expenditures of
the income.
This paper focuses at sentiment at an individual level, therefore neither the previously mentioned balance score nor the consumer sentiment index (CSI) can be used. Therefore, the following two methods are developed to created a sentiment index the household level. First, we adapt eq. (1) in the following manner. For PP(very positive) and P(Positive) answers we respectively assign the value 2 and 1
2. Similarly, we assign the values of − 1
2 and -2 to M(negative) and MM(very negative)
answers4. Let i be a household index and consider question j (j = 1, .., 5), equation 1 becomes:
Bijt= (P Pijt+
1
2Pijt) − ( 1
2Mijt+ M Mijt), j = 1, ..., 5 The first sentiment index is calculated as follows:
sent 1it= 5
X
j=1
Bijt (3)
Where j denotes the question, i the household and t the year.
For the second sentiment index, of every individual question all answers are normalized so that they have mean 0 zero and a standard deviation of 1.
Zijt= e Xijt− fXj ˜ sj where f Xj= T P t=1 Nt P i=1 g Xijt T P t=1 Nt
4The questions INKNORM and INKEVEN have only three possible answers. For a positive answers we used 3 4
and ˜ sj = v u u u u t 1 ( T P t=1 Nt− 1) T X t=1 Nt X i=1 ( eXijt− fXj)2 !
The second individual sentiment index is the sum of all normalized values of all five questions.
sent 2it= 5
X
i=1
Zit (4)
−1 −.5 0 .5 1 1.5 2000 2005 2010 2015 year
sent1 sent2 sent3
sent1 stockholders sent2 stockholders sent3 stockholders
Figure 1: Sentiment trend
3.2
Financial wealth and control variables
As the control variables the following characteristics variables are used: highest level of completed education, number of children in the household,house owner, net disposable income and age. For the financial asset holding variables the following calculations are used:
f inancial wealth = Tchecking+ Tsaving+ Tmutualf und+ Tbonds+ Tshares (5)
risk ratio = risky assets
f inancial wealth (7)
Where Tchecking is the total financial wealth in checking accounts. Tsaving is the total financial
wealth in saving accounts, Tmutualsf und is the total financial wealth in mutual funds,Tbonds is the
total financial wealth in bonds and Tshares is the total financial wealth in shares.
See fig. 2 for the percentage of stock ownership of households over the years. As can be seen in the graph there is an uptrend is household stock market participation till the year 2009. After which the ownership drops dramatically. Probably because of falling stock prices due to the financial crisis. .16 .18 .2 .22 .24 .26 percentage 2000 2005 2010 2015 year
3.3
Limitations
Not all sentiment question are answered by all the participants, apppendix A2 shows the total responses per question and appendix A3 the number of answered sentiment questions. Also, some households consist of several members who fill out the survey. For computational purposes we only use the responses of the head of the households. If the only respondent is not the head of the household, the answers of the partner is used. Furthermore, the survey is only held among dutch household. As Guiso et al. (2003) show there are large differences between nations when it comes to holding risky assets.
3.4
Descriptive statistics
Table 1: Summary statistics
Variable Obs Mean Std. Dev. Obs Mean Std. Dev. Obs Mean Std. Dev.
full sample estimate sample stock owner sample
sent 1 16967 0.608 2.135 15128 0.620 2.117 3697 1.327 2.214 sent 2 16967 0.000 3.540 15128 0.021 3.527 3697 1.128 3.409 sent 3 16967 0.000 3.608 15128 0.011 3.597 3697 1.143 3.486 inknorm 16094 1.958 0.261 14396 1.958 0.261 3611 1.985 0.213 inkrond 16967 3.418 0.871 15128 3.419 0.864 3697 3.742 0.837 finsitu 16962 3.412 0.995 15124 3.415 0.990 3696 3.694 0.943 inkeven 16962 2.172 0.725 15124 2.174 0.723 3696 2.339 0.744 ink25a 16959 2.995 0.871 15121 3.005 0.866 3695 3.117 0.916 age 17665 53.625 15.105 15128 54.400 14.975 3791 56.807 14.239 no. children 17666 0.599 1.008 15128 0.561 0.981 3791 0.519 0.992 male 17666 0.702 0.458 15128 0.707 0.455 3791 0.800 0.400 education 17648 4.866 1.510 15128 4.852 1.512 3787 5.271 1.466 disp. hh. income 17666 30.876 26.912 15128 30.925 27.603 3791 35.592 26.697 house owner 17666 0.696 0.460 15128 0.694 0.461 3791 0.818 0.386
wealth in checking ac-counts
15623 2484.154 7842.373 15128 2489.330 7893.675 3791 3427.336 9961.745
wealth in mutual funds 15623 6521.677 36565.310 15128 6616.777 36982.500 3791 26876.330 70454.720
wealth in bonds 15623 2121.250 21841.080 15128 2136.783 21964.580 3791 6537.877 39478.810
wealth in shares 15623 3694.213 30264.660 15128 3732.706 30649.900 3791 15224.130 59998.940
wealth in savings 15623 20454.550 71068.610 15128 20720.960 72006.360 3791 36813.210 93268.670
4
Econometric Model
In this section the model to estimate our hypotheses is developed. First, the ownership of risky assets on a binary scale is tested. Therefor a dummy variable risky assets is created, which is equal to 1 if either the amount of financial wealth in mutual funds or the amount in stock is larger than 0.
4.1
Binary Probit model for risky asset ownership
Hypothesis 1 is tested following Hurd et al. (2011), let yit = risky assets and let pit be the
restricted to the interval [0,1]. The model takes the following form:
pit= P r(yit= 1|sentit, xit) = Φ(sentitβ1+ x0itβ2+ γt) (8)
Where Φ(.) is the cumulative distribution function (cdf) of the standard normal distribution, sentitour constructed sentiment index and xita vector of explanatory control variables, including
a constant, as described in the previous section.
The relationship between the ownership of risky assets and household sentiment will be mod-eled by means of a probit model. Clustered standard errors account for unobserved heterogeneity between the households. Clustered standard errors are consistent in the presence of arbitrary het-eroskedasticity and within-households serial correlation (Rothstein, 2005). It is expected that the signs of the sentiment indicating variables will be positive. Thus meaning that the sentiment of a household will have a positive effect on the probability of owning risky assets. One of the draw-back of the estimation of a probit model is that only the signs of the coefficients can be usefully interpreted. In order to gain more from the data marginal effects are measured using the following methods:
For continuous variables the average marginal effect of sentiment is measured as follows: 1 N N X i=1 ∂yit ∂xijt = 1 N T T X t=1 N X i=1 ∂P r(yit= 1|sentit, xit, ) ∂xijt = 1 N N X i=1 Φ(x0itβi)βj (9)
where φ(.) is the value of the standard normal distribution. The marginal effect for a dummy variable xijt is equal to:
1 nT T X N X
4.2
OLS model for risky asset ratio
Secondly, a model for our second hypothesis is developed. Testing for the ratio of risky assets as described in 2. Let risk ratioitbe this ratio that household i holds in year t. The relation between
the risky asset ratio and household sentiment, among owners of risky assets, will be modeled by means of the a pooled OLS model. Using clustered standard errors to account for unobserved heterogeneity between the households.
risk ratioit= β1sentit+ β2xit+ it (11)
Where risk ratioitis the ratio of risky assets over all assets, sentitis the same sentiment indexes
for households as described and used previously. We test this model against the same vector x0itof control variables.
5
Results
be found in appendix A6. With the highest log-likelihood, the probit model using sentiment 2 and time dummies (column 2 ) seems to be the best fit of the model.
5.1
Probit model
All the statements below assume ceteris paribus. Results of the models are very similar. In all the models it can be observed that the variable of interest, the sentiment variable, is positive and highly significant. It can be concluded that household sentiment does positively increases a households probability of holding risky assets. As for the age variable, it is positive and significant, age also increase the odds of holding risky assets as found by Ameriks and Zeldes (2004). The second order effect of age is negative and significant, meaning that this effect diminishes as households grow older. Furthermore, males are also more likely to hold risky assets than females, as found by Hurd et al. (2011) and others. This can attributed to the fact that males in general are more risk loving. Looking at the highest level of completed education, we see see that compared to only com-pleted primary education, university schooled households are most likely to hold risky assets. The disposable income affects the probability of holding risky assets also in a positive significant way. Lastly we see that house owners are significantly more likely to hold risky assets over tenants.
5.1.1 marginal effects
sentiment index 3 is very similar to sentiment index 2 with a marginal effect coefficient of 1.49% (see appendix A6). With every year older a household is on average is about 1.2% more likely to hold risky assets. Males are on average around 7.3% more likely to hold risky assets compared to female. Holding a university degree or a vocational college degree respectively increases on average the likelihood of holding risky assets with 10.4% and 21.2%. Disposable income also has a statistical significant impact, however this is hardly economically significant. Owning a house does increase the probability of holding risky assets with around 9%.
5.1.2 marginal effect graphs
For some of the variables the marginal effects of the probit model at different levels are calculated. This was done for the three sentiment variables figs. 5 to 7 , the education variable fig. 4 (with a step for every possible answer to the questionnaire) and for the age variable fig. 3 with steps of 5 years. For age and education we the probit model using sentiment index 1 is used(eq. (3)). The graphical representations can be seen below.
0 .2 .4 .6 .8 1 Pr(Stockdummy) 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 age
Adjusted Predictions with 95% CIs
Figure 3: age .1 .2 .3 .4 Pr(Stockdummy)
kinderg/primary educ university education
highest level of education completed
Adjusted Predictions with 95% CIs
Figure 4: education 0 .2 .4 .6 .8 Pr(Stockdummy) −15 −13 −11 −9 −7 −5 −3 −1 1 3 5 7 9 11 13 15 sentiment1
Adjusted Predictions with 95% CIs
Figure 5: sent1 0 .2 .4 .6 Pr(Stockdummy) −15 −13 −11 −9 −7 −5 −3 −1 1 3 5 7 9 11 13 15 sentiment2
Adjusted Predictions with 95% CIs
Figure 6: sent2 0 .2 .4 .6 Pr(Stockdummy) −15 −13 −11 −9 −7 −5 −3 −1 1 3 5 7 9 11 13 15 sentiment3
Adjusted Predictions with 95% CIs
Looking at marginal effects of different levels of completed education the following can be stated. The overall trend is that a higher level of completed education relates to a higher probability of holding risky assets. One might argue that this due to higher financial literacy. However, the level of education is an imperfect measure for financial literacy (van Rooij et al., 2011).
Lastly, the marginal effects graphs for the three sentiment indexes. In all cases, when the sentiment index increases there is an increasing effect in the probability of owning risky assets. Where the plotted line of sentiment index 2 and 3 show a more linear function. The sentiment index 1 shows a more s-shaped function.
Table 2: Sentiment 1, with and without years (1) (2) (3) (4) sentiment1 0.0919∗∗∗ 0.0883∗∗∗ (0.012) (0.012) sentiment2 0.0517∗∗∗ 0.0501∗∗∗ (0.007) (0.007) age 0.0386∗∗ 0.0383∗∗ 0.0389∗∗ 0.0386∗∗ (0.012) (0.012) (0.012) (0.012) age squared -0.000257∗ -0.000269∗ -0.000262∗ -0.000273∗ (0.000) (0.000) (0.000) (0.000)
number of children in the house-hold 0.00934 0.00636 0.00482 0.00223 (0.029) (0.029) (0.029) (0.028) male 0.252∗∗∗ 0.267∗∗∗ 0.254∗∗∗ 0.267∗∗∗ (0.063) (0.062) (0.063) (0.062) kinderg/primary educ 0 0 0 0 (.) (.) (.) (.)
vmbo [pre-vocational educ] 0.0391 0.0252 0.0306 0.0170
(0.156) (0.155) (0.157) (0.156)
Havo/vwo [pre-univ educ] 0.472∗∗ 0.466∗∗ 0.464∗∗ 0.458∗∗
(0.165) (0.163) (0.166) (0.165)
senior vocational or apprentice syst 0.251 0.235 0.240 0.224 (0.162) (0.161) (0.163) (0.162) vocational colleges 0.427∗∗ 0.405∗∗ 0.419∗∗ 0.397∗ (0.155) (0.154) (0.156) (0.155) university education 0.746∗∗∗ 0.718∗∗∗ 0.751∗∗∗ 0.723∗∗∗ (0.161) (0.160) (0.163) (0.161)
net disp hh income () 0.00120 0.000962 0.00127 0.00103
(0.001) (0.001) (0.001) (0.001)
home owner 0.324∗∗∗ 0.312∗∗∗ 0.326∗∗∗ 0.315∗∗∗
(0.065) (0.065) (0.065) (0.064)
constant -2.574∗∗∗ -2.801∗∗∗ -2.511∗∗∗ -2.741∗∗∗
(0.345) (0.341) (0.344) (0.341)
year dummy yes no yes no
Observations 15128 15128 15128 15128
Table 3: Pooled OLS for sentiment index 1,2 and 3 (1) (2) (3) sentiment1 -0.0277∗∗∗ (0.008) sentiment2 -0.0223∗∗ (0.007) sentiment3 -0.0220∗∗ (0.007) age 0.00997 0.00959 0.00953 (0.007) (0.007) (0.007) age squared -0.0000815 -0.0000773 -0.0000770 (0.000) (0.000) (0.000)
number of children in the household -0.00897 -0.0105 -0.0106
(0.022) (0.022) (0.022)
male 0.0443 0.0486 0.0488
(0.030) (0.030) (0.031)
vmbo [pre-vocational educ] -0.0776 -0.0748 -0.0745
(0.073) (0.071) (0.071)
Havo/vwo [pre-univ educ] -0.0530 -0.0512 -0.0514
(0.072) (0.071) (0.071)
senior vocational or apprentice syst -0.0869 -0.0857 -0.0858
(0.075) (0.073) (0.073)
vocational colleges -0.0568 -0.0540 -0.0538
(0.070) (0.069) (0.069)
university education -0.0531 -0.0528 -0.0525
(0.070) (0.068) (0.068)
net disp hh income 0.000384 0.000474 0.000476
(0.000) (0.000) (0.000)
home owner 0.0295 0.0331 0.0333
(0.032) (0.032) (0.032)
constant 0.213 0.195 0.206
(0.231) (0.228) (0.229)
year dummy yes yes yes
Observations 3693 3693 3693
R-squared 0.0174 0.0225 0.0229
Errors clustered at the household level in parentheses
5.2
Risk ratio model
The second model tested is the model with as a dependent variable as ratio of risky assets over total financial wealth (see eq. (7)), using eq. (11). The regression results for all three sentiment indexes are presented in table 3. Not many significant coefficient are observed. Only the sentiment indexes are negatively significant related to the risk ration. This means that, ceteris paribus, when sentiment increases the allocation of financial wealth becomes less risky. This against our expectation as presented in hypothesis 2. Where we expected that higher levels of sentiment would positively influence the risk ratio rather than negative.
5.3
Robustness
Aforementioned results do seem promising but lack dept in order to be truly confirmed. By testing the probit model using a Mundlak approach see Mundlak (1978) we can distinguish between the within and between variation of the sentiment indexes of households. As can be see in appendix A7 the household averages of age and sentiment index do absorb a lot of the explanatory power of the sentiment indexes. None of the indexes is significant anymore. Therefore, personal deviations from a households mean do barely explain the probabilities of holding risky assets. The significance of the sentiment indexes is mainly explained due to between variation, rather than within variation.
It could be the case that an underlying characteristics also influences riskiness of asset allocation. For example general risk aversion. Therefore the following two questions we used from the DHS for further research:
money
Both questions are indicated on a scale of 1 to 7. Where 1 indicates ”totally disagree” and 7 indicates ”totally agree”
When including these question as dummy variables in our regression and using Mundlak fixed effects a different effect than in table 2 or appendix A7 is observed. The results are presented in appendix A8. Again, all the three sentiment indexes are not significant, the mundlak household averages are. However, the risk aversion questions remain significant even when we control for their average. This can indicate that a risk aversion proxy using these question might be a better explanatory variable for understanding the household stock holding puzzle. Nonetheless, this is outside the scope of this paper and can be the subject of further research.
6
Conclusion
In this paper we develop two hypothesis for testing the relation between sentiment levels on a micro scale and the stock holding puzzle. Using household data from the Dutch Household Survey (DHS) we construct three different sentiment indexes. For all three indexes we find evidence that household sentiment has a positive effect on the probability that a household holds risky assets.
References
Adams, F. G. (1964). Consumer attitudes, buying plans, and purchases of durable goods: A principal components, time series approach. The Review of Economics and Statistics, 347– 355.
Ameriks, J., & Zeldes, S. P. (2004). How do household portfolio shares vary with age (Tech. Rep.). working paper, Columbia University.
Amromin, G., & Sharpe, S. A. (2009). Expectations of risk and return among household investors: Are their sharpe ratios countercyclical? Available at SSRN 1327134 .
Barber, B. M., & Odean, T. (2008). All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. Review of Financial Studies, 21 (2), 785–818.
Bodie, Z., Merton, R. C., & Samuelson, W. F. (1992). Labor supply flexibility and portfolio choice in a life cycle model. Journal of economic dynamics and control , 16 (3), 427–449.
Boswijk, H. P., Hommes, C. H., & Manzan, S. (2007). Behavioral heterogeneity in stock prices. Journal of Economic dynamics and control , 31 (6), 1938–1970.
Brown, G. W., & Cliff, M. T. (2005). Investor sentiment and asset valuation*. The Journal of Business, 78 (2), 405–440.
Charoenrook, A. (2005). Does sentiment matter. Unpublished working paper. Vanderbilt University. Curtin, R. (2007). Consumer sentiment surveys. Journal of business cycle measurement and
analysis, 2007 (1), 7–42.
De Long, J. B., Shleifer, A., Summers, L. H., & Waldmann, R. J. (1990). Noise trader risk in financial markets. Journal of political Economy, 703–738.
the health and retirement study. Journal of the European Economic Association, 5 (2-3), 369–379.
Golinelli, R., & Parigi, G. (2004). Consumer sentiment and economic activity. Journal of Business Cycle Measurement and Analysis, 2004 (2), 147–170.
Guiso, L., Haliassos, M., & Jappelli, T. (2003). Household stockholding in europe: where do we stand and where do we go? Economic Policy, 18 (36), 123–170.
Haliassos, M., & Bertaut, C. C. (1995). Why do so few hold stocks? the economic Journal , 1110–1129.
Hurd, M., van Rooij, M., & Winter, J. (2011). Stock market expectations of dutch households. Journal of Applied Econometrics, 26 (3), 416–436.
Hymans, S. H., Ackley, G., & Juster, F. T. (1970). Consumer durable spending: explanation and prediction. Brookings Papers on Economic Activity, 173–206.
Jansen, W. J., & Nahuis, N. J. (2003). The stock market and consumer confidence: European evidence. Economics Letters, 79 (1), 89–98.
Keane, M. P., & Runkle, D. E. (1990). Testing the rationality of price forecasts: New evidence from panel data. The American Economic Review , 714–735.
Kimball, M. S. (1993). Standard risk aversion. Econometrica: Journal of the Econometric Society, 589–611.
King, M. A., & Leape, J. I. (1998). Wealth and portfolio composition: Theory and evidence. Journal of Public Economics, 69 (2), 155–193.
Lemmon, M., & Portniaguina, E. (2006). Consumer confidence and asset prices: Some empirical evidence. Review of Financial Studies, 19 (4), 1499–1529.
the American Statistical Association, 58 (304), 899–917.
Mundlak, Y. (1978). On the pooling of time series and cross section data. Econometrica: journal of the Econometric Society, 69–85.
of Michigan Survey Research Center, U. (2015). Surveys of consumers (Tech. Rep.). Author. Otoo, M. W. (1999). Consumer sentiment and the stock market.
Qiu, L., & Welch, I. (2004). Investor sentiment measures (Tech. Rep.). National Bureau of Economic Research.
Rothstein, J. (2005). Does competition among public schools benefit students and taxpayers? a comment on hoxby (2000) (Tech. Rep.). National Bureau of Economic Research.
Shapiro, H. T. (1972). The index of consumer sentiment and economic forecasting: A reappraisal. Human Behaviour in Economic Affairs: Essays in Honour of George Katona, New York, Elsevier Scientific Publishing.
Shum, P., & Faig, M. (2006). What explains household stock holdings? Journal of Banking & Finance, 30 (9), 2579–2597.
Souleles, N. S. (2004). Expectations, heterogeneous forecast errors, and consumption: Micro evi-dence from the michigan consumer sentiment surveys. Journal of Money, Credit and Banking, 39–72.
Appendix A
.1
Michigan ICS questions
For the calculation of the Michigan ICS the following question are asked:
• X1 = ”We are interested in how people are getting along financially these days. Would you
say that you (and your family living there) are better off or worse off financially than you were a year ago?” Possible answers: 1) Better know 2) Same 3) Worse
• X2 = ”Now looking ahead–do you think that a year from now you (and your family living
there) will be better off financially, or worse off, or just about the same as now?” Possible answers: 1) Will be better of 2) Same 3) Will be worse off
• X3= ”Now turning to business conditions in the country as a whole–do you think that during
the next twelve months we’ll have good times financially, or bad times, or what? Possible answers: 1) Good times 2) Good with qualifications 3) Pro-Con 4) Bad with qualifications 5) Bad times
• X4 = ”Looking ahead, which would you say is more likely–that in the country as a whole
we’ll have continuous good times during the next five years or so, or that we will have periods of widespread unemployment or depression, or what?” Possible answers: 1) Better 2) Worse 3) Same
• X5 = ”About the big things people buy for their homes–such as furniture, a refrigerator,
Table A2: Number of responses to the 6 separate sentiment questions
Question Number of responses
inknorm 13392 inkrond 14126 finsitu 14123 inkeven 14123 ink25a 14122 geluk 13986 total households 17,666
Table A3: Number of answered sentiment questions
Total number of answered sentiment questions No.
0 2,881.0 1 660.0 2 1.0 3 1.0 4 68.0 5 1,395.0 6 12,660.0 total 17,666.0
Table A4: Number of households per year that hold assets
year households households with risky assets percentage with risky assets
Table A5: Marginal effects sentiment index 1 and 2, with and without years (1) (2) (3) (4) sent 1 0.0273∗∗∗ 0.0264∗∗∗ (0.004) (0.004) sent 2 0.0154∗∗∗ 0.0150∗∗∗ (0.002) (0.002) age 0.0115∗∗ 0.0115∗∗ 0.0116∗∗ 0.0115∗∗ (0.004) (0.004) (0.004) (0.004) age squared -0.0000764∗ -0.0000805∗ -0.0000779∗ -0.0000818∗ (0.000) (0.000) (0.000) (0.000) number of children 0.00278 0.00190 0.00143 0.000666 (0.009) (0.009) (0.009) (0.009) male 0.0720∗∗∗ 0.0762∗∗∗ 0.0723∗∗∗ 0.0764∗∗∗ (0.017) (0.017) (0.017) (0.017)
vmbo [pre-vocational educ] 0.00868 0.00570 0.00680 0.00386
(0.034) (0.035) (0.035) (0.035)
Havo/vwo [pre-univ educ] 0.129∗∗ 0.130∗∗ 0.127∗∗ 0.128∗∗
(0.041) (0.041) (0.041) (0.042)
senior vocational or apprentice syst 0.0620 0.0589 0.0594 0.0564 (0.037) (0.038) (0.038) (0.038) vocational colleges 0.114∗∗ 0.110∗∗ 0.113∗∗ 0.108∗∗ (0.036) (0.036) (0.036) (0.037) university education 0.226∗∗∗ 0.219∗∗∗ 0.229∗∗∗ 0.222∗∗∗ (0.041) (0.041) (0.041) (0.041)
net disp hh income 0.000358 0.000288 0.000377 0.000307
(0.000) (0.000) (0.000) (0.000)
home owner 0.0914∗∗∗ 0.0889∗∗∗ 0.0922∗∗∗ 0.0897∗∗∗
(0.017) (0.017) (0.017) (0.017)
year dummy yes no yes no
Errors clustered at the household level in parentheses
Table A6: Probit model sentiment 3(1) plus marginal effects (1) (2) main sent 3 0.0500∗∗∗ 0.0149∗∗∗ (0.007) (0.002) age 0.0389∗∗ 0.0116∗∗ (0.012) (0.004) age squared -0.000261∗ -0.0000778∗ (0.000) (0.000) number of children 0.00454 0.00135 (0.029) (0.009) male 0.254∗∗∗ 0.0724∗∗∗ (0.063) (0.017)
vmbo [pre-vocational educ] 0.0303 0.00675
(0.157) (0.035)
Havo/vwo [pre-univ educ] 0.464∗∗ 0.127∗∗
(0.166) (0.041) senior vocational or apprentice syst 0.240 0.0594 (0.163) (0.038)
vocational colleges 0.419∗∗ 0.113∗∗
(0.156) (0.036)
university education 0.752∗∗∗ 0.229∗∗∗
(0.163) (0.041)
net disp hh income 0.00128 0.000380
(0.001) (0.000) home owner 0.327∗∗∗ 0.0924∗∗∗ (0.065) (0.017) constant -2.531∗∗∗ (0.344) Observations 15128 15128 Log-likelihood -7665.4 Pseudo R-squared 0.0883
Errors clustered at the household level in parentheses
Table A7: Mundlack approach probit (1) (2) (3) sent 1 0.00948 (0.008) sent 2 0.00516 (0.005) sent 3 0.00444 (0.005) age 0.0168 0.0166 0.0120 (0.016) (0.016) (0.016) age squared -0.000254∗ -0.000261∗ -0.000262∗ (0.000) (0.000) (0.000) number of children 0.0226 0.0180 0.0178 (0.029) (0.029) (0.029) male 0.238∗∗∗ 0.238∗∗∗ 0.239∗∗∗ (0.064) (0.064) (0.064)
vmbo [pre-vocational educ] -0.000816 -0.0139 -0.0149
(0.161) (0.162) (0.162)
Havo/vwo [pre-univ educ] 0.387∗ 0.373∗ 0.372∗
(0.188) (0.188) (0.188)
senior vocational or apprentice syst 0.139 0.123 0.121
(0.200) (0.200) (0.200)
vocational colleges 0.256 0.242 0.238
(0.222) (0.221) (0.220)
university education 0.523∗ 0.526∗ 0.522∗
(0.262) (0.261) (0.261)
net disp hh income 0.00105 0.00112 0.00112
Table A8: Mundlack approach probit plus risk attitude (1) (2) (3) sent 1 0.038 (0.009) sent 2 0.031 (0.005) sent 3 0.026 (0.005) age 0.881 (0.017) 0.886 (0.017) 0.738 (0.017) age squared -1.408∗∗ (0.000) -1.445∗∗ (0.000) -1.447∗∗ (0.000)
number of children in the household 0.059 (0.030) 0.051 (0.030) 0.051 (0.030)
male -0.067 (0.068) -0.069 (0.068) -0.067 (0.068)
vmbo [pre-vocational educ] 0.010 (0.164) -0.003 (0.165) -0.004 (0.165)
Havo/vwo [pre-univ educ] 0.204 (0.194) 0.195 (0.194) 0.194 (0.194)
senior vocational or apprentice syst 0.189 (0.214) 0.178 (0.213) 0.176 (0.213)
vocational colleges 0.277 (0.237) 0.267 (0.237) 0.264 (0.237)
university education 0.345 (0.284) 0.349 (0.284) 0.346 (0.283)
net disp hh income () 0.011 (0.001) 0.015 (0.001) 0.015 (0.001)
