Inspired by recent critiques on the empirical validity of prospect theory, I'm excited to share new paper explicitly testing dynamic predictions of PT.

Punchline: Using field data + pre-registered experiments, find very strong support for PT. (Thread)

https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3600583

To summarize, we:
1. Empirically identify new form of dynamic inconsistency specific to risky choice.
1. Run formal horserace between models, find strong support for specific, non-obvious predictions of PT in dynamic settings.
3. Explain some puzzles in risky choice.

Prospect Theory makes specific prediction on dynamic inconsistency in planned versus actual behavior:

1. Begin taking risk as part of "loss-exit" strategy: plan to keep taking risk after gains and stop after losses

2. Actual behavior is opposite: chase losses + cut gains early

To identify this dynamic inconsistency, we first use unique data set of investors (N=190,000), which gives access to their planned strategy (loss/gain limits) and actual behavior for *every* position opened.

Majority plan to sell after small losses, but keep asset after gains.

Actual behavior is *opposite*: the vast majority keep losing positions for way longer than planned, and sell gains much earlier.

In our setting, this deviation--- i.e. dynamic inconsistency --- results in significant financial losses.

Next, ran two pre-registered experiments to test not just qualitative but quantitative predictions of PT.

Simple setting with multiple treatments: People face sequence of gambles, after finding out outcome of one, presented with opportunity for another. Can stop anytime.

First, do people accept the first gamble as part of "loss-exit" plan?

Yes! 81% of people enter with a "loss-exit plan in mind!

Difference is substantial: Plan to take four times as many gambles after gains than after a losses.

What does behavior look like once they start seeing losses/gains?

The exact opposite! Even in first round, chance of exiting after gain is *triple* that of exit after loss.

This identifies predicted dynamic inconsistency. Further, plans quantitatively match theory predictions.

Why is this important? Explains large discrepancy in risk-taking between one-shot and dynamic settings.

Prior work shows people refuse one positive EV gamble, but accept gamble as part of sequence, even when EV is negative

Implies risk-aversion and risk-seeking in same person!

Prospect theory, and our data, rationalize this discrepancy by looking at how the ability to condition behavior on prior outcomes---i.e. ability to plan---changes skew over final earnings.

"Loss-Exit" plans generate much more positive skew over final outcomes than what's available in one-shot settings.

Probability weighting---a central feature of PT---makes this strategy much more attractive than a one-shot gamble.

We find evidence for this as well: when offered 50/50, zero EV gambles either in isolation or as part of sequence, people much more likely to take it in latter case---as part of ``loss-exit'' plan---than former case

Importantly, ability to plan does not change EV: still zero!

So, in dynamic settings people take risk they would otherwise refuse. What happens after?

They completely deviate from their plans! In both of our settings, the deviations lead to financial losses.

Moreover, dynamic inconsistency implies welfare costs apart from financial ones

Paper has lots of other results, re demand for commitment and specific policy implications.

Would love to hear your thoughts!

Fin.

Just to add: this is not meant to be a critique of the critiques. Rather, hopefully this highlights the predictive power of the theory in an empirically relevant setting.