Open research commons · Issue 01

A public notebook for serious curiosity

School of Thought is open to everyone. Publish the question you could not stop thinking about, show your work, and give other curious people something they can examine, cite and build on.

Featured research

Issue 01
MarketsMarket microstructure and options
ProbabilityBayesian inference and rationality

A journal anyone can contribute to

Think of it as an open repository for ideas. Contributors publish original research, make their reasoning inspectable and invite others to question, cite or extend the work. These first two papers are featured examples, not the boundary of what belongs here.

All papers

Research article · Quantitative finance

Inside the Ban: A Quantitative Autopsy of Jane Street's Trading Tactics in India

A quantitative reconstruction of the two-legged index manipulation strategy alleged by India's securities regulator, including index mechanics, option repricing, impact cost and net profit.

Abstract

In July 2025, the Securities and Exchange Board of India accused Jane Street Group of manipulating index-linked derivatives through a two-legged strategy that distorted price discovery. This paper offers a forensic breakdown of the alleged trading loop, the resulting market dislocations, the regulator's interpretation and a compact quantitative model of how a loss-making cash leg can support a larger derivatives profit.

1. Context and strategy

The action included an asset freeze totaling 48.4 billion rupees, about $570 million in the paper's source material. The central allegation is not simply that an index moved, but that activity in selected cash securities was coordinated with a pre-positioned derivatives book.

Leg 1 · Cash marketAggressive trading in selected, less liquid index constituents mechanically moves the NIFTY level. This description applies to the targeted subset, not the whole cash market.
Leg 2 · Derivatives marketA long-delta position in liquid NIFTY index options appreciates as the underlying index shifts and convexity takes hold.

NIFTY 50 is a benchmark index representing 50 large-cap companies listed on India's National Stock Exchange. Because the index is built from weighted constituent prices, a sufficiently forceful move in a subset of stocks can transmit into the reported index level.

2. Mathematical model

2.1 Index mechanics

Let It denote the index value at time t. For n constituent prices Si,t with weights wi, the index and its induced change are:

It = ∑i=1n wiSi,t
ΔI = It+1 - It = ∑i=1n wiΔSi

Aggressive buying during a low-volume period puts upward pressure on the selected constituent prices. The weighted sum transmits that pressure into the index.

2.2 Option pricing impact

For a trader long at-the-money NIFTY calls, the mark-to-market value of the option leg can be approximated by a second-order Taylor expansion:

PnLoption = Δ · ΔI + ½Γ · (ΔI)2Delta measures first-order sensitivity. Gamma measures curvature.

The gamma term makes the payoff nonlinear. Close to expiry, at-the-money options can have especially high gamma, so a small underlying move may create a disproportionately large change in option value.

2.3 Impact cost and net profit

The cash leg pays spread and slippage. The paper models this cost as a linear function of the absolute index move:

PnLcash = -κ · |ΔI|
PnLnet = PnLoption + PnLcash

In the paper's illustrative eight-basis-point move, the option leg gains 8.4 crore rupees while the cash leg loses 2.1 crore rupees, leaving a net 6.3 crore rupees. One crore equals 10 million, so the example net is 63 million rupees.

₹8.4 CrOption-leg gain
₹2.1 CrCash-leg loss
₹6.3 CrIllustrative net profit
8 bpsAssumed index move

The core implication is mechanical: a deliberately loss-making leg can be rational within the combined book when the convex derivatives payoff grows faster than the cost of moving the cash market.

3. Execution sequence

  1. Position in optionsAccumulate long-delta, at-the-money calls while they remain comparatively inexpensive. Near expiry, gamma and convexity are especially important.
  2. Engineer an index moveBuy a basket of selected constituents with low float, wide spreads or high price impact per unit of volume.
  3. Capture repricingThe higher index level lifts the calls. A simultaneous increase in quoted implied volatility can add a second source of appreciation.
  4. Exit and flattenSell the options at higher prices, then reverse the cash trades. The index may revert, but the combined book can finish with no exposure and a realized profit.

Repeated across expiries and instruments, this loop may produce a recognizable strategy fingerprint even when each cycle is small or embedded in high-frequency execution.

4. Market reaction

The paper records a cluster of market signals around the episode. These figures are reproduced from the original article:

-0.83%NIFTY intraday move on 18 July 2025
+14.6%Change in ATM NIFTY implied volatility
+242%Mid-cap cash volume versus 20-day average
-1.2 to -2.5Widening in 25-delta risk reversal, vols

The paper identifies the Securities Contracts (Regulation) Act of 1956 and the Prohibition of Fraudulent and Unfair Trade Practices framework of 2003 as relevant to the regulatory action. Penalties described include the 48.4 billion rupee asset freeze, bank-account restrictions and a prohibition on market access.

6. How the pattern was detected

6.1 Surveillance triggers

Unexplained volume surges, repeated early-session activity and price moves without matching news or analyst revisions provided the first layer of evidence. What could look isolated in one market becomes more suspicious when compared across venues and timestamps.

6.2 Cross-market correlation

SEBI analysts compared cash trades with index-derivatives order books. The paper describes a sequence in which short-term index lifts were preceded by trade bursts in selected cash securities, while long-delta options had been accumulated shortly beforehand.

6.3 Strategy fingerprints

  • Recurring delta-positive option purchases in the 15 minutes before an index lift
  • Rapid reversal of constituent trades after option positions were sold
  • Linkage through nested sub-UCC codes and offshore FPI counterparty structures
“The artificial lift in index value was not the result of fundamental discovery, but was rather a synthetic outcome of coordinated intraday activity.”

The broader lesson is methodological. Surveillance of modern market manipulation must reconstruct an economic strategy across instruments, not merely identify an unusual order in isolation.

7. Sources

  1. Bloomberg, India Bars Jane Street From Accessing Its Securities Market
  2. Bloomberg, Jane Street and SEBI explainer
  3. Securities and Exchange Board of India, interim order PDF
  4. NSDL, foreign investors data
Article note: This page preserves the author's analysis and illustrative model. It is not investment advice. Allegations and regulatory findings should be read alongside the cited primary order.
Article 001 · Sandra Cai · 2025Next paper
All papers

Research article · Probability and decisions

Theorem of Wisdom: Bayes' Theorem as the Most Possible Rational Way of Making Decisions

A practical argument for replacing one-shot intuition with priors, likelihoods and an updateable view of the world.

Abstract

People make thousands of decisions, but intuition alone does not guarantee a rational outcome. Bayes' theorem offers a quantitative way to combine what was plausible before an observation with how well each possible cause explains the new evidence. This paper develops that idea through familiar scenarios and argues for Bayesian updating as a lean, repeatable approach to everyday reasoning.

1. Introduction

Bayes' theorem belongs to conditional probability. It asks how the probability of one event changes when another event is already known to have occurred. If event A has no influence on event B, the events are independent. Otherwise, their joint probability is:

P(A ∩ B) = P(A) · P(B | A)

Rearranging this relationship gives a basic conditional probability:

P(B | A) = P(A ∩ B) / P(A)Valid when P(A) > 0

Daily life is full of informal inference: reconstructing an event from its traces, deciding whether clouds imply rain, or reading another person's expression. The danger is collapsing a field of possible causes into the first explanation that feels vivid. A probabilistic mindset keeps several hypotheses alive long enough to compare them.

2. Worked scenario: who is more likely to be drunk?

Suppose Ben is drunk in 9 of 10 drinking occasions and Jerry is drunk in 1 of 10. Assume, before observing anyone, that either person is equally likely to be the one in view.

Likelihood for BenP(drunk | Ben) = 0.9
Likelihood for JerryP(drunk | Jerry) = 0.1
Prior for BenP(Ben) = 0.5
Prior for JerryP(Jerry) = 0.5

The probability of observing drunkenness is the weighted sum of the two routes:

P(drunk) = 0.9 · 0.5 + 0.1 · 0.5 = 0.5

Bayes' theorem then reverses the direction of the question:

P(Ben | drunk) = (0.9 · 0.5) / 0.5 = 90%
P(Jerry | drunk) = (0.1 · 0.5) / 0.5 = 10%

The conclusion is not produced by likelihood alone. It emerges from likelihood and prior probability working together.

3. Worked scenario: is the elevator going to crash?

A sudden, severe change in elevator speed can feel like strong evidence for catastrophe. A crash would explain the sensation very well, but explanatory power is only the likelihood term. It says nothing about how often elevator crashes occur before this observation is made.

The paper compares two hypotheses: a crash, and a functioning but heavily used elevator that moves unevenly. Using its stated assumptions:

Crash score = 1.00 · 9.52 × 10-8
Overuse score = 0.10 · 0.90 = 0.09

Under those assumptions, the overuse explanation receives roughly 94,538 times the unnormalized posterior weight of the crash explanation. The dramatic hypothesis has greater apparent explanatory force, yet its tiny prior keeps its posterior probability low.

Modeling note: The figures here reproduce the paper's illustrative assumptions. In a real safety decision, priors and likelihoods should come from reliable operating and incident data, and unusual mechanical behavior should still be reported.

4. Interpreting the theorem

For a reason that might explain an observed phenomenon, Bayes' theorem can be written as:

P(reason | phenomenon) = P(reason) · P(phenomenon | reason) / P(phenomenon)
PriorHow plausible the reason was before the new observation.
LikelihoodHow well that reason predicts the observation.
EvidenceThe overall probability of the observation across the competing reasons.
PosteriorThe updated probability of the reason after seeing the observation.

When comparing hypotheses for the same observation, the evidence term is shared. Ranking the unnormalized products of prior and likelihood is therefore enough to see which explanation leads.

Posterior probability ∝ prior probability × likelihood

5. A practical Bayesian method

  1. Observe the phenomenonDescribe what was actually seen, without smuggling an explanation into the observation.
  2. List possible causesKeep several live hypotheses, including ordinary ones and the possibility that the list is incomplete.
  3. Estimate priorsAsk how common each cause was before this new evidence appeared.
  4. Estimate likelihoodsAsk how probable the observation would be if each cause were true.
  5. Update and compareMultiply prior by likelihood, normalize when exact posterior probabilities matter, then compare.
  6. Repeat with new evidenceTreat belief as updateable. A rational conclusion is a current estimate, not a permanent identity.

6. Summary

Bayesian reasoning provides a disciplined alternative to choosing whichever explanation most dramatically fits a new fact. The theorem requires both base rates and explanatory power. That matters in science, where hypotheses should gain or lose credibility as evidence arrives, and in ordinary life, where striking explanations often feel more likely than they are.

The practical value of Bayes' theorem is not certainty. It is a repeatable way to become less wrong as information accumulates.

The method also reveals why disagreement can persist. People may assign very different priors from different experiences, then update rationally from the same evidence and still land apart. Making those priors explicit creates a better starting point for genuine discussion.

Article note: This page follows the ideas, examples and calculations in the author's original paper, with light editorial restructuring for the web edition.
Article 002 · Sandra Cai · 2025Read the finance paper

Contribute

Put your strange question on the record

You do not need an institution, a credential or a conventional topic. You need a clear question, honest methods and work that another person can follow.

Submission entry

Complete the cover note, then use your device's share sheet to send the paper and details together. Nothing is uploaded when you select a file.

A concise summary is enough. Aim for 100 to 300 words.
PDF is preferred. You can use a public link instead.
or
Use a stable link that readers can open without requesting access.

Your browser will ask which email or sharing app to use. If file sharing is unavailable, a cover note downloads for you to attach with the paper manually.

About the commons

Research belongs to curious people

School of Thought is a shared place for rigorous investigations that begin outside a lab, a company or a syllabus.

Open by default

Anyone can contribute. A paper might come from a student, an engineer, a collector, a cook, a neighborhood historian or a person who simply kept pulling on a question.

The standard is not status. It is whether the work makes its question, evidence and reasoning visible.

Built to be continued

Each paper is a starting point rather than a final word. Readers should be able to cite it, challenge it, reproduce it and develop the next version of the idea.

Like an open repository, the value grows when contributors show their work and leave a clear trail for the next curious person.

What belongs here

Careful work on an unexpected topic is welcome: a quantitative autopsy of a market event, a model of an everyday decision, an experiment in a kitchen, an archive of a tiny subculture, or an investigation that has no obvious academic home.

Read the guidelines and contribute