Cary Frydman

Behavioral economist and professor @USCMarshall

[Taha Choukhmane @TahaChoukhmane]

🧵 1/11 Half of U.S. adults already use AI for financial advice: what advice are they getting?

New paper: we ask people to write prompts seeking financial advice from LLMs & simulate the lifetime impact of following that advice (w/ @timdesilva, @wdwlin &
@AkuzawaMatthew)

[Cary Frydman @caryfrydman]

@malling @dividendology Why wait for the dividends? Just automate a sell 2% of each position every quarter, make your own dividend

[Cary Frydman @caryfrydman]

@lw_alena @RRholes Very cool paper. This would probably boost the measured pass through from beliefs to choice as well

[NBER @nberpubs]

An experimental test of efficient coding shows that experience tends to crowd out the effect of descriptive information on optimal allocation of coding resources, from @caryfrydman and @lawrence_j_jin https://t.co/hmF1IW9427

[Emir Efendić @EmirEfen]

Been away from socials, but want to mention a new project that I'm really proud of.

In this pre-print we present evidence for a simple solution on how to find accurate answers to open-ended questions: select the person fastest to respond (thread ⬇️)
https://t.co/69WOQPxOGw

[Cary Frydman @caryfrydman]

@BenBushong Congrats!!

[Cary Frydman @caryfrydman]

Overall message: reported beliefs drive valuations, but by too little. Valuations are optimally anchored given cognitive constraints

[Cary Frydman @caryfrydman]

Now out in the JF with Constantin Charles @c_charles_23 and Mete Kilic @mkilicLA! In an experiment, we find that subjects report asset valuations that are far too insensitive to their expectations -- as in field data among @Vanguard investors

https://t.co/ekgbRwsdGJ

[Cary Frydman @caryfrydman]

One other key finding is that the passthrough from beliefs to actions depends heavily on the *type* of beliefs. For a given distribution of beliefs, valuation is way more sensitive to objective beliefs compared to subjective beliefs

[Tim de Silva @timdesilva]

💡Implication: the growth of target date funds as defaults might be good for investors by helping align choices & preferences (more to come on this)

Note that if non-participation was driven by loss-aversion, as some literature suggests ,this would not necessarily be the case!

[Cary Frydman @caryfrydman]

This is amazing. The bias that language models have in computing expected value depends on the training data. And when the training data matches the distribution of real-world stimuli, you get biases that look like those in human choice data

[Zhu Jian-Qiao @JQ_Zhu]

10/10 Our result suggests that language models have internalized these classical patterns reported in economics and psychology. And these behaviors likely emerge from learning to calculate expected values.

[Cary Frydman @caryfrydman]

@Bellmanequation @deadmilkmen While you’re at it, tag their record label headed by @CFCamerer !

[Jessica Flack @C4COMPUTATION]

RIP Danny Kahneman

Kahneman recently told me in a discussion about collective intelligence that if he had at least another decade to live, he'd found a department to study the role of noise in human behavior.

Of course noise--a phenomenon often with counterintuitive effects--is well studied in collective behavior, information theory, dynamical systems + engineering, etc., but remains somewhat underappreciated in the social sciences.

One my favorite concepts of Kahneman's + Tversky's is the simulation heuristic (Judgement Under Uncertainty, 1982, Chapter 14)--the idea that we natively assign higher subjective probabilities to scenarios that are "easier" to consciously run forward or play out as a counterfactual. So many questions about computation, intelligence, + consciousness are raised by this simple idea.

https://t.co/ceQVKg9Mlb

A broadly important open micro-macro question that is finally getting traction is whether heuristics + biases (like the above) are as K+T and their disciples argue necessary corrections to standard economic theory or would (in some compressed form) follow naturally + simply as predictions from the equations *if* we had the right model. For more on this idea, see the work of Ole Peters.

[Cary Frydman @caryfrydman]

@CFCamerer It’s basically a mini Caltech HSS PhD, except colder

[Rafael Polania @RafaPolania]

Take a look at our Review just published in @TrendsCognSci
Rationality, preferences, and emotions with biological constraints: it all starts from our senses

Free access to our article (online and PDF) for about 40 days!! don't miss it 👇
https://t.co/kRTqBvdDOe

[Cary Frydman @caryfrydman]

Interesting new explanation for behavior in coordination games using machine learning. Data on context-dependent coordination from a recent paper with @NunnariSalvo

[NBER @nberpubs]

Variational autoencoders can model imprecision in the way that people coordinate their behavior in laboratory experiments, and adapt to new environments, from Guy Aridor, Rava Azeredo da Silveira, and Michael Woodford https://t.co/yOC4Wlw1An

[Cary Frydman @caryfrydman]

@KrishnaSrinii @SandroAmbuehl 💯 need to factor in that if the unlikely event is true, it can move people’s beliefs a lot

[Robert Metcalfe @RDMetcalfe]

Ok folks, here is the job ad for our Empirical IO Assistant Professor position and how to apply:
https://t.co/Fca7SsM8G6

We'll be moving fast so please encourage all of your students to apply quickly.