Meetings

Oct 7 2025

• What Holger is working on; anything on LLMs; what he thinks of my idea; logistics (Zurich visit, Brazil experiment era)
• Holger: could even be inviting a couple of people to talk, does not have to be a full workshop. Wait until he comes to discuss it.

Oct 16 2025

• Soccer idea: rulebook says harm is the rule for foul, but actual rule is whether the person touched the ball
• Comments on categorical norms paper:
  • Holger thought continuous signal is what matters, not continuous actions
  • They show there is an equilibrium with continuous actions
• Holger's question: is it only about information structure? What about people agreeing to disagree? People consistently favor their own group when deciding whether norms are abided by
  • Me: could be explained by combination of different preferences + signal structure (require stronger signal if in-group since will face less social sanctions for type I vs type II errors)
• Holger: must be literature on cartels with private monitoring that is relevant
• From Sagua & Wolitzky JPE 2018: first attempt to investigate whether transparency facilitates collusion

Oct 17 2025

• Using lawyers connected to judges apparently also used in the US:
  • Northern district Texas bankruptcy court scandal: the key lawyer in the "local partner firm" of the main law firm channeling bankruptcy litigation there was having an affair with the judge
  • In a patent litigation scandal the relative of the judge was engaging in patent litigation in that same town
• SCOTUS votes are very predictable — just goes in the direction of judge's political ideology, irrespective of judicial philosophy
• Holger's idea: one judge watches another judge to (i) learn whether the other judge is defecting (and thus whether to pull the grim trigger) or (ii) learn what the law is, given that the other judge is thought to not have defected
  • (ii) is why judges influence each other
• Could the theory just be that if one judge does not uphold the law, then others might not as well (Kandori style)?
  • My idea: in such a game, you would be especially worried about defecting if the other judge has opposite preferences (e.g., political ideology) from you (e.g., the peer in an appeal panel)
  • This might have implications for having rotation or not between panels
• Holger thinks the most interesting is appeal panels, since trial judges can always be appealed, and as long as appeal judges decide according to the law they get incentives

Oct 20 2025

Discussion topics:

• Why punishing at 50% threshold in civil cases efficient?
• Duncan Kennedy "A Critique of Adjudication" recommended: need to do work to get away from accepted meaning; you cannot just cite another judge
• HL Hart "no vehicles in the park" example
• Dworkin chain novel theory: next judge writes the next chapter, but must be coherent with what came before

Signal and equilibrium discussion:

• Your signal can always be signal + epsilon. Bounded equilibrium. Or by mixing.
• What you think is the law is allowed to change over time. You can learn from the other judges — then I take the average of your and my signal
• Over time they will converge, and punishment zone will increase
• Don't converge on a point, since still some private info
• Personal preferences — friends of workers or business owners?
• Cannot agree to disagree
• Some noise in the facts of each law + preferences
• Rule of evidence is about the jury

Institutional examples:

• Product liability: can sue not only seller but the producer; no contract with producer, so a tort had to be invented; courts said "look at this case about sheep, and the principle that applies is this..."
• Free speech has shifted over time
• Freedom of broadcasting in Germany: interpreted it as prohibiting private; if part of private corporation then journalists not free
• Fiduciary duties: content of those duties is a continuous thing
• Large linguistic literature on vagueness of language
• Tax lawyers can say two things in same conversation without seeing a contradiction: "this transaction circumvents the tax law" / "but not caught by the tax law"

My thinking:

• Idea that language has different meaning for people makes sense. "Reasonable" might mean something different to people, while "horse" is unlikely to have different meaning.
• How can we measure the indeterminacy of words?
• Can we think of a concrete example where everything applies? Constitutional provision about executive power in the US? An example where law is very binding and clear? — "Horses are not allowed on the highway"
• Judges must distinguish case from precedents?
• There could be a myriad of different facts that could have happened in the world; each case is unique; but world might be discrete in certain ways that could be explained by words, such as "horse" — a well-defined empirical category with borderline cases very unlikely
• Typically "things" can be easily described; actions are harder; abstract things like causes, effects, intent even harder

Thoughts on precedents:

• If a lower court judge decides differently from a controlling precedent, they must distinguish it — explain why the facts or legal issues differ
• Failing to do so is grounds for reversal on appeal
• Example: if the Supreme Court has held that "X constitutes a search under the Fourth Amendment," a district court can't simply rule the opposite

Oct 21 2025

• Holger wants a model where people are uncertain about what the law is and look at other judges' decisions to learn about it
• In my framework, we could probably have judges observing X+e in all past cases + the decision
• If they assume others follow the law, then they can learn other judges' signals (i.e., their reading of past cases)

Thoughts:

• We seem to agree that law is a mapping from the space of possible facts to binary decisions (liable/not liable)
• Past cases show realizations; each new case is unique
• But cases might also introduce general principles guiding how to extrapolate to new cases
• Judges might disagree about how to make such extrapolations
• Maybe they see a vector X1, X2, ... and have a model f(X1, X2, ...) that projects this to the unit line
• Say there is a "true" fact X, but judges see X+e (when trying to project a case to the line)

Oct 21 email comment on model (from Holger)

Reactions:

1. Your base model is essentially Anthony Niblett's.
2. Extension:
   • The error epsilon can't be additive and homoscedastic if cases are distributed on a finite interval, as in your uniform assumption (technical point, but not unimportant for the base setup)
   • What do you mean by judges lying about their signal?
     • Do only they see the case, X? In that case, isn't there complete unraveling immediately — what's stopping the judge from representing their case in whatever way they want? Perhaps they want to set a precedent, but what good is the precedent if the other judges also just represent the case in any way they want?
     • A better and more realistic version is that other judges also see the case, but with bigger error.

Comment on correlation of signals (see email)

Total payoff in your model is 1-delta(q+r), where r is the probability of both signals being wrong and q the probability that only one (particular one) is wrong. Let's call the marginal probability of any one signal being wrong pi, and the correlation between the two rho. Then:

• q = pi(1-pi)(1-rho)
• r = pi^2 + rho*pi(1-pi)

Thus, q+r = pi. That means the total payoff only depends on the (marginal) probability of a wrong signal, not on the signal correlation. And the payoff is decreasing with that probability, as one would expect.

Dec 9 2025

• Holger thinks it is good that binding force of precedent arrives endogenously
• My counterexample of prisão após segunda instância: clearly lower court judges complied with STF because of a norm to do so, not because they changed their belief about "the law"
  • Holger's counterargument: yes, but in a hierarchical system appeals give judges incentives to abide by higher courts. That is a different mechanism. It is more interesting to study how judges at the same level (e.g., circuit court judges) bind each other.
• Holger will read up on and share Kornhauser articles
• What Holger wants from the model:
  • Where is the binding force of law coming from. What determines its strength?
  • Precedents bind in equilibrium

Dec 15 2025

Multidimensional model:

• Cases lie in a multidimensional space
• There could be many theories of the law that fit these points in space equally well
• There could also be a preference for functional forms (e.g., linear/simple preferred — as in the Hand rule)
• Judges can select their preferred model among those that fit prior cases in ways that align with their preferences
• Past cases that do not fit the model are interpreted as "errors"
  • Example: Roe v Wade was interpreted as an error by current court
• All judges can be seen as applying "the law" in a reasonable way, even though ending on different conclusions

More thoughts:

• Maybe relates to "narrative" literature?

Dec 15 prep notes

"Second, we show, somewhat counterintuitively, that judges are more likely to apply a consistent rule when the bench is highly polarized than when it is ideologically cohesive." (from stare decisis log-rolls paper) — cool idea

• Why do judges care how other judges decide?
• Judges might also collude on a rule to prevent litigation
• Judges could refer to past cases to avoid criticism ("don't blame me")
• Would also save time

To me:

• Many reasons why stare decisis is valuable

• Question is how it is sustained / how to increase the cost of deviating

• Nice to have contested vs settled regions — but these could be settled by supreme court right? (Answer: but that is vertical stare decisis)

Jan 13 2026

• Case only binds in neighbourhood? Or decaying in force?
• Monotonicity idea nice way of formalizing holding
• Lot of binary categories. Is or is not police interrogation
  • Binary: was there a contract or no? No degree

BI visit / workshop planning

Relevant law people

• Morten Kinander
• Runar Hilleren Lie
• Sjur Swensen Ellingsæter

Relevant finance people

• Bogdan

People who could attend miniworkshop

1. Øyvind Stiansen
2. Daniel Naurin (UiO, judicial politics)
3. Runar Hilleren Lie — asylum case project
4. Malcolm
5. Bergen people
6. PhD / Postdoc researchers connected to PluriCourts or IKR (Oslo Faculty of Law)
7. Jørn Øyrehagen Sunde, Jon Christian Nordrum — collaborators in "Quality of reasoning in automated judicial decisions" project

My thoughts on invitees

• Arnaud Philippe
• Wessel Wijtvliet (KU Leuven)
• Nordrasil people for Copenhagen
• Urška Šadl (Copenhagen)
• Authors of Springer book on judicial decision-making (2022)
• Piotr Bystranowski — edited book on judicial decision-making

ChatGPT suggestions

Europe:

• Giacomo Ponzetto (CREI/Barcelona GSE) — already suggested by Holger; formal models of judicial behavior and legal interpretation
• Edoardo Grillo (University of Padova) — formal political theory and judicial decision-making; legal ambiguity and statutory interpretation
• Matthias Dahm (University of Nottingham) — judicial discretion and institutional design
• Johannes Wachs (CEU Vienna) — network science + NLP applied to court decisions (Germany, Austria)
• Jan-Emmanuel De Neve (Oxford) — behavioral economics, subjective judgment
• Daniel Chen (Toulouse) — LLMs and NLP for judge behavior, causal inference

US:

• Ryan Copus (University of Missouri) — legal tech + empirical analysis of judicial decision-making; built LLM tools for court analytics
• Jonas Lerman (Yale Law School) — AI in legal interpretation
• David Engstrom (Stanford) — RegLab; applied AI/ML for legal reasoning; institutions and discretion
• Jonathan Masur (University of Chicago) — judicial behavior, law-and-economics and behavioral angle

Funding for accommodation

• Fulbright: application deadline in September for next year
• NFR international researcher recruitment scheme (but seems to be for permanent, not short-term)
• Finansmarkedsfondet
• ASF (American-Scandinavian Foundation): deadline Nov 25 for upcoming year