 The starting point of this study is the question, how do households invest their savings? Why is it the case that some households take up a lot of risk and invest their savings in stocks or funds or other risky assets, but other households put all their money on a savings account and don't take any risk? The latter households, however, forego a substantial amount of earnings, especially if they could invest their money for a long time horizon. The factor I'm especially interested in my study is subjective beliefs of households about the development of the stock market. That is, maybe the differences can be explained by the fact that some households are just more optimistic about the returns they can make than other households. This idea is at odds to most economic and traditional models, because those models assume that all households have rational expectations and that differences in observed behavior is only driven by differences in preferences, most importantly risk preferences, and by differences in the wealth level of these households. Both of these factors actually matter for these choices, but a large share of observed heterogeneity remains unexplained by these factors, and this is where my study comes in. The question, if beliefs matter for financial decisions, has important implications. On the one hand, if the decision of households can be mostly explained by preferences, this means that households make optimal decisions for themselves. On the other hand, if beliefs matter here, this could imply that if beliefs are too extreme, for example that they are too pessimistic compared to previous returns on the stock market. This would mean that information campaigns or coachings could improve the decisions of households and can help these households to make better choices. To answer my research question, I need two important components of data. First, I need information about the portfolio of households. I get this information from administrative records based on tax data in the Netherlands. In this data, I can observe what share of the portfolio of households is invested in risky assets, like for example stocks, and what share of the portfolio is invested in safe assets. This measure is arguably much more precisely measured in this administrative data compared to, for example, survey data where households are asked for their assets themselves. The second component is then the expectations about stock market development, and I get this information from a household survey. This household survey is a probability sample of the Dutch population. Most importantly, it can be linked on the individual level to the admin data. In this survey data, I can make use of two elicitations of beliefs. Both of them are incentivized. So I make use of these two data components, put them together, and then I'm able to look at the relation of those two measures. My analysis proceeds in two steps. First, I look at the cross-sectional relation, that is, I look only at one point in time. What I find there is that indeed, households that are more optimistic about the stock market also tend to hold more risky assets in their portfolio. Importantly, this relation still holds if I control for a rich set of control variables that include risk aversion and numeracy. Both of them have been shown to be relevant for portfolio choice of households. The effect size is roughly half of the effect size for risk aversion. Then I proceed to my second analysis, and in there I make use of the second elicitation of beliefs. Because half a year after the first elicitation, households were again asked for their beliefs about the future development of the stock market. What I find there is that those households that became more optimistic over this half a year increased their share of risky assets in their portfolio over a similar timeframe. My paper is not the first one looking at the relation of beliefs and portfolio choice. Three features make my paper stand out. First, I'm making use of this administrative data that is arguably of a higher quality than survey data. Secondly, I look at a broad population sample. That is, I'm not only looking at investors or at rich households, but I'm getting a broad picture of the Dutch population. Thirdly, I make use of this second elicitation of beliefs. Because the problem if I only look at cross-sectional analysis is that the correlation that I observed there could be actually driven by a third factor, for example, personality characteristics. If my personality characteristics are, for example, related to both my investment behavior and to my beliefs about the stock market, then I would observe a correlation there without actually being a causal link between these two factors. I can address this concern with my dynamic analysis, because by showing that also changes over time are related to changes over time in the portfolio choice, I can show that it's not a time invariant third variable like, for example, personality characteristic that is driving the results. All together, I can confirm that expectations are an important component of portfolio choice of households. I would like to mention three potential directions of future research. First, it would be interesting to experimentally vary beliefs of households and see how this affects actual portfolio choices. This would make the identification of the effect I'm interested in even cleaner and would then allow to identify a causal link between the two concepts. The second direction of future research I would like to mention is to go deeper into the interplay of beliefs, risk preferences, but also preferences about ambiguity. And to further understand the complex interaction between these three concepts and to better understand then how households actually make these kind of financial decisions. Thirdly, in a more general sense, my study shows that beliefs matter and that beliefs can be reliably measured in surveys. This would mean that more so than it's actually already done right now. It will be interesting in the future to look at choices of households in other contexts like, for example, education choices, various labor market decisions and then to see if those might be also driven by expectations of these households.