Investing money? Looking for a new job?
Often, we do not know the probabilities of potential outcomes (aka ambiguity).
Yet, we know much less about ambiguity attitudes, compared to risk attitudes.
New WP (joint with @econ_hmg and Axel Wogrolly) tries to change that.
Results:
For at least 3/4 of the population, it makes a big difference whether probabilities are known or not.
Attitudes are heterogeneous in the cross-section, but stable over time.
Attitudes predict portfolio choice -- much higher explanatory power than similar studies for risk pref.
Based on repeated elicitations of ambiguity attitudes (11.000 person x wave observations) in a probability sample of the NL (🇳🇱 ) we examine:
- Distribution of attitudes in three dimensions
- Stability over time and across domains
- Relation to pot. determinants and outcomes
We make use of the LISS panel:
- true probability sample of the Dutch population
- takes great care that every selected person can attend (e.g. provide computer if necessary)
- includes a rich set of control variables and data on assets
Subjects make a series of binary choices between:
A) a bet on the development of the stock market (ambiguous option)
B) a random draw with know probability (risky option)
Collect 6 biannual waves of data. In one wave additional application: climate change
We estimate a structural choice model, based on three parameters:
- (average) ambiguity aversion
- likelihood insensitivity/perceived level of ambiguity
- error parameter
Subjects are on average ambiguity seeking for low probability events and ambiguity
averse for mid to high probability events.
But: Heterogeneity is high.
Parameters are highly stable across time (esp. when accounting for measurement error).
Ambiguity aversion highly stable across domains. Level of ambiguity less so (-> fits to interpretation as cognitive component)
-> Overall, stability very similar to risk attitudes
We use ML-clustering techniques (k-means) to obtain 4 preference types
1. "Near SEU":
- Choices close to subjective expected utility maximization: low level of ambiguity, ambiguity neutral
- Highest eduction, highest numeracy
- Highest income and wealth
2. "Ambiguity averse" and 3) "Ambiguity seeking"
- Perceive high levels of ambiguity -- either averse or seeking towards it
4. "High noise":
- Inconsistent choices
- Older subjects, lowest numeracy
Ambiguity types are highly predictive of portfolio risk -- even when controlling for a rich set of control variables (like risk aversion, numeracy)
Robustness check: same results for administrative asset data based on tax records
-> Results validate estimated parameters
Two main take-aways:
1. Ambiguity attitudes are important to understand decisions under ambiguity (beyond risk attitudes)
2. Ambiguity attitudes are measured more precisely by:
- Explicitly taking into account decision errors
- Accounting for joint distribution of parameters (clustering)
Find the WP here: https://docs.iza.org/dp15712.pdf
Replication Package: https://github.com/ChristianZimpelmann/replication-ambig-beliefs
If you like this thread, feel free to share the first toot: https://econtwitter.net/@c_zimpelmann/109659877966918499
Christian Zimpelmann