Université d'Orléans · Sciences Économiques · 29 Nov 2004
Four chapters on how a currency price is made.Three of them run in your browser.
A dissertation on exchange rate dynamics and currency crises, rebuilt in 2026 as laboratories you can operate — the cascade simulator, the agent-based market, the estimators on today's series. From the original manuscript. No new research, no rework.
The spine
One mechanism, running through four chapters
Information does not arrive evenly. Everything else in the dissertation follows from that — including the moment a fixed parity breaks.
A multiplicative cascade
The construction physicists built for turbulence, splitting intensity unevenly all the way down.
Information intensity
Bursts and lulls in how fast news actually reaches a decentralised market.
Random trading time
Trades arrive against that intensity, not on a calendar grid — the chapter's core departure.
The parity breaks
Not when a fundamental crosses a threshold — when enough people revise at once.
Multiplicative cascadesPoisson arrivals in business timeLux–Marchesi switching under random trading timeGPH & local Whittle d̂(q)Structure functions ζ(q)
The work, as instruments
Three laboratories you can operate
Every one of them takes your input and re-runs in front of you. Nothing is precomputed and nothing phones home.
Chapter One Live
Towards a multifractal paradigm of stochastic volatility
Fat tails, volatility that clusters, memory that changes with the power you measure — almost every familiar feature of returns follows from one assumption about how information arrives.
- Drag intermittency λ² and watch excess kurtosis climb on the live tiles
- Stop the cascade partway down, and see which horizons were carrying the volatility
- Switch between six views — cascade, information flow, returns, memory, scaling
- Read the original chapter beside the simulator: forty equations, four tables
9 controls · 6 views · 4,096 observations · seeded · Alcatel 1991–2001, as reported
Chapter Two Live
Agent-based financial market simulation
A market made of people who disagree. Two chartist camps and a fundamentalist camp, each agent switching when someone else's strategy is doing better — and, unlike almost every simulation of its day, trading time is random rather than a grid.
- Push the herding dial α₁ up and watch the price detach from the fundamental
- Send good or bad news into the crowd and watch the three camps change size
- Race ten random clocks against the calendar grid — the chapter's core departure
- Step the market maker one tick at a time, through ten events written out in full
4 dials · 4 presets · 200 traders live · 2,500 simulated days at seed 20030601
Chapter Three Live
The same estimators, on today's markets
Chapter One found persistence in volatility that falls away as the power measured rises, while persistence in trading volume barely moves. This runs the same estimators over ten years of markets that did not exist in that sample, and draws both curves on one axis.
- Switch between five daily series, 2016–2026: MSFT, BTC, EUR/USD, US 10y, Brent
- Read measured d̂(q) against Chapter One's Alcatel table, with a ±2 s.e. band
- Change the definition of volatility — |r|, r², rolling σ — and watch the curve move
- Compare GPH with local Whittle, and ζ(q) past the simulator's k = 64 ceiling
5 series · 2016–2026 · GPH + local Whittle · q = 0.25…4 on the chapter's own grid
Chapter Four In preparation
Speculative attacks on a fixed exchange rate market
The case the other three chapters were built for: a fixed parity, a central bank defending it, and a population of speculators who revise together. Being rebuilt from the manuscript on the same terms as the rest.
- Frontier AI
- 1 minuteto analyse the chapter
- Me
- 1 yearto master and write it
2003 · 2026
A personal note, 23 years later
In September 2026, OpenAI announced a resolution to the Navier–Stokes Millennium Problem. The news immediately took me back to my PhD.
More than twenty years ago, while doing my PhD in finance, I was working with a research group in France, composed mainly of physicists (CNRS Econophysique). They were exploring how ideas developed to understand turbulence in fluid, earthquakes or avalanches could also be applied to financial market crashes and volatility. This was completely new territory for me. And it was fascinating.
I took two of their instruments into my PhD. Volatility I modelled with multifractal processes — the multiplicative cascade built to describe turbulence, carried across to the uneven way information arrives in a market. I also built 2 agent-based simulations (yes, like the AI agents of 2026, although they were “deterministic” at that time):
- a market of traders who keep changing their minds;
- a speculative attack on a fixed exchange rate.
That is the other standard instrument physicists use on complex systems — you do not write the aggregate behaviour into the equations, you let it emerge from a population of “agents” interacting within a simulation.
When I returned to the thesis last week, I knew the digital files were lost. The PDF was gone. The code was gone. All I had was the printed copy. I took pictures of every page and digitized the full document. I then gave the first chapter to GPT. In less than 1 minute, it analysed the research that took me 1 year to master and write. It was a shock! But its conclusion was reassuring: for some work written more than two decades ago, it was not bad. Some passages had aged surprisingly well.
So I decided to put the old work online. With few coding agents, I rebuilt the models, tests and simulations as interactive laboratories. A few hours later, this website is now live.
The interface is new. The research is not.
What I found fascinating: rebuilding this work in no time gave me a concrete example of the acceleration now taking place in research. Powerful AI systems radically compress the time needed between an idea and a working experiment. Work that once required months or years can now be reconstructed, tested and shared in hours.
Each lab here remains deliberately faithful to the original work. No rework. Nothing is quietly corrected to make my younger self look smarter.
Bucharest, 15 September 2026
The intuition
Why I approached currency crises this way
I started this research effort on currency crises with a first Dissertation done during my Master. Macroeconomics. Focus on the fundamentals behind these crises. Interesting, but I couldn't understand the timing of these crises. I quickly felt I was in a deadend. I realized I needed to give up my focus on macroeconomics to understand what — concretely — was happening in the FX markets. I needed to understand how a financial market works, and why/how it can create some crashes. Market Efficiency: if prices move because news arrives, and news reaches everyone at once and is read in much the same way, then the volatility we actually observe in currency markets is far too large. You can find the intraday spikes around macro-announcements — but they are a small part of the total. Most of the volatility was being produced by something else than public information.
The FX market is decentralised by construction: there is no aggregate order flow everyone can observe. So what a trader learns about everyone else, they learn from price and volume themselves. Other participants are not noise around the fundamental — they are part of what you are trading on. That is also why technical analysis dominates short-horizon forecasting there, whatever one thinks of it.
So I stopped treating the representative investor on a regular clock as the starting point (once again, the main flaws of the Efficient Market Hypothesis). Take it away and you need to say what replaces it, which is the whole dissertation: information that arrives in bursts, agents who revise their convictions by watching what is working for other people, and a trading time that runs fast and slow instead of ticking constantly.
A fixed exchange rate does not break because a fundamental crossed a threshold on a particular Tuesday. It breaks because enough participants revise at once, each partly because the others are revising — a herd that is individually rational and collectively catastrophic. A representative agent cannot even state that problem. A population that switches strategy, in a market where the only signal about everyone else is the price, can.
That is why the emerging-market application at the end is not an afterthought bolted onto the theory. It is the case the theory was built for.
A note on the rebuilds. Each laboratory is written from its chapter's own equations, not from its published figures. Where the manuscript is ambiguous, silent, or missing pages, the implementation says so on the page and names the reading it took. Nothing is quietly corrected and nothing is modernised.
Université d'Orléans
Thèse
présentée
à l'Université d'Orléans
pour obtenir le grade de
Docteur de l'Université d'Orléans
Discipline : Sciences Économiques
par
Nicolas Boitout
Modélisation de la dynamique des taux de change avec application aux marchés émergents
Soutenue publiquement le 29 novembre 2004
Provenance
From one printed copy
The digital files were lost — no PDF, no code. Every page of the surviving printed copy was photographed and digitised, and every laboratory here is written from its chapter's own equations rather than from its published figures.
- Degree
- Docteur de l'Université d'Orléans
- Discipline
- Sciences Économiques
- Defended
- 29 November 2004
- Director
- Prof. Cyrille Piatecki
N. Boitout & L. Ureche-Rangau, “Towards a Multifractal Paradigm of Stochastic Volatility?”, International Journal of Theoretical and Applied Finance 7(7), 823–851, 2004 · 10.1142/S0219024904002736
Bibliography
What it was built on
The manuscript's own reference list, restandardized to APA 7th and repaired where the printed copy was damaged. Working papers that were later published carry their final journal; every DOI that exists resolves. Nothing has been added to it.
- References
- 353
- DOIs resolved
- 290
- Span
- 1951 → 2010
Defence and jury
Université d'Orléans · 29 November 2004
- Emmanuel AcarHead of Foreign Exchange Risk Management, Bank of America House, London
- Gilbert ColletazProfessor, Université d'Orléans
- Thomas LuxProfessor, University of Kiel · rapporteur
- Valérie MignonProfessor, Université Paris X Nanterre · rapporteur
- Cyrille PiateckiProfessor, Université d'Orléans · research director
- Gilles TeyssièreScientific Director, NBG Banque, Paris