Raisa

Introduction#

Main RQ: Personal experiences $\Rightarrow$ Housing price expectations

In contrast to previous research, in this study the local market of individuals is not defined by geography but by density, since both housing price observations and the exposure of different generations to these prices may depend on the density of the residential area.

  • Lifetime (global) experiences
  • Local experiences (limited to area)
  • Direct experiences (recent)

Main challenge: compare individuals who just had the experience to the individuals who did not.

  • This way, you can get an unbiased estimate of the causal effect of those experiences
  • If not, it might be conflated with e.g. region-specific (or age-specific) factors, even after controlling for observables

Questions#

Since I have rewritten the introduction and literature review to fit a substantially changed research question, there is not a specific part that I think is most important right now. My main question is: am I on the right track?

I think it looks coherent! There should maybe be a fourth section in chapter 2, briefly reviewing the influence of other factors (that are not the focus of your research) on expectation formation (“fundamentals” and other factors).

My idea is to calculate a weighted average of annual real housing price growth experienced over the adult life for each age in the sample. Since the study relies on yearly data, this only leads to 50 unique observations of weighted average annual real housing price growth when using data from one year. Could this be resolved by using data for multiple years? Because the study focuses on the boom in the Dutch housing market, this would imply the years 2013-2019 and lead to seven times the 50 unique observations = 350 observations. Or are there other ways to improve the measurement of this variable? See also the attached file which shows how the weighted average will be calculated. I intend to estimate the regression for different levels of x to see which one best fits the data.

I am not sure whether I understand this well, but I don’t think it suits your RQ very well: you are not talking about the influence of average housing price growth. I think it is better to exploit discontinuities in exposure between different cohorts (some cohorts saw rising/falling, some cohorts saw only rising, etc.) This can be exploited using something similar to Regression discontinuity or Regression kink.

I interchangeably use the words age, younger and older people, and generations, does this lead to confusion or is it okay?

I think that’s okay.

Note: I have discussed using the LISS Panel or DHS data with Dr. Paaso a few times. I chose to use the DHS data, because the LISS Panel has only asked respondents for their housing price expectations in 2011. This is outside of the context of the current Dutch housing boom and still in a year during which the market was on the decline after the crisis of 2007-2008. Therefore, I prefer the DHS data which collects information on housing price expectations every year.

I agree that this suits your RQ better (you do need to have some kind of panel/cohort structure in your data).

Methodology and Data#

I think all hypotheses can be tested using a panel model (estimated by OLS).

H1: Age is negatively related to housing price expectations, this relationship is mediated by national housing price growth experienced during the adult life.

H2: Age is negatively related to housing price expectations, this relationship is mediated by the density of the residential area in which an individual lives.

H3: Age is negatively related to housing price expectations, this relationship is mediated by participation in the housing market as a first-time home buyer.

However, the first hypothesis is a bit more difficult. Other studies on aggregate lifetime experiences rely either on the age cohorts themselves to test the effects or extensive econometric modelling. The study by Knüpfer et al. (2017) approaches it differently by looking at changes in unemployment shares for 817 combinations of region and occupation.

I would say: do you have access to data on at what age people buy their first homes (historically)? Then you can exploit that information to test whether there is a discontinuity / kink (Regression Kink Design ) in the relationship between age and housing price expectations

There is one methodological issue I’m worried about. I think hypothesis 2 and 3 can simply be tested with OLS regression and interaction terms between age cohorts and dummy variables for the level of density of the residential area and for the type of participation in the housing market.

Hypothesis 2 can be tested by employing this model (so you don’t need dummies I think):

$$ Exp_{i} = \alpha + \beta \cdot \text{Age}_{i} + \gamma \cdot \text{Density}_{i} + \\
\delta \cdot \text{Age x Density}_{i} + \phi \cdot \text{Controls}_{i} + \epsilon_i $$

But a disadvantage of this design is that it does not take into account that people choose where to live. Hence, it might be better to employ matching-estimators to find similar observations of people that are treated (live in a high density) and people that are untreated (live in a low density), and compute the average treatment effect from these matches.

H3: Do you mean recent experience as first-time home buyer? This is very difficult to test, because of the reverse causality (expectations influence your decision to buy a home). I think it will not generally be possible to find exogenous variation in participation on the housing market. That means you can test this hypothesis, but it will be very difficult to claim it reflects accurately the true impact of participation on expectations.