Published with the course instructor's agreement; all market data is anonymised.
Baptiste Cristofari · NHH · FIN11 Trading and Market Microstructure · Fall 2026
Over 5 classroom sessions I traded 3 different market structures on screen and 1, open outcry, by voice. This page covers each session, compares how closely prices followed value, and my thoughts on it. Reading time: ≈8 minutes.
Contents: What I learnt · Session 1 · Session 2 · Session 3 · Open outcry · Session 4 · Why prices followed value · What I built · Repository
| Session | Date | Market | My role |
|---|---|---|---|
| 1 | 18 Aug | One asset, limit order book, 25,000 price cap | Trader |
| 2 | 25 Aug | One stock, limit order book, private news | Not in this session |
| 3 | 7 Sep | Two stocks, limit order books, three roles | News trader |
| Outcry | 8 Sep | Trading floor simulation, no screen | Broker, buying 30,000 |
| 4 | 14 Sep | Two stocks, dealers' quotes only | Asset exchanger |
- Market structure decides whether prices match the fundamental value. See why prices followed value in Session 3 and not the others.
- A market that is not efficient can ignore the fundamental value for hours. In Session 2, 95% of trades went through at more than 2x the fundamental value the news traders were estimating.
- Wait for a good price, not for the clock. In Session 4 I sold STOCK1 at up to 4x the fundamental value while spreads were tight, then left most of my STOCK2 to the end and had to buy it at 2x the fundamental value.
- Without a book, trading first is costly. In the open outcry I bought too high in one of the room's first deals, and that single trade dropped my final rank from 4th to 9th out of 13.
18 August 2026 · FTS Web Trader · 50 minutes
Setup. Everyone traded the same invented asset, bottles. Each of us started with 20 or 60 bottles and no cash, and a bottle still held at the bell was worth nothing:
total gain = cash + realised utility + 0 × bottles
The session was a race to sell the endowment before the bell. The server also rejected any price above 25,000, and that cap ended up shaping the whole session.
Trade prices, bid and ask, and the bottles I held, over the 50 minutes
The bid and ask above come from the trades themselves: a trade started by a buyer went through at the ask, and one started by a seller at the bid. The platform's order log cannot give a cleaner answer, because it keeps orders nobody could fill: one ask at 2 sat in it all session while bottles printed at 25,000.
What happened. It was the first session, so nobody knew how to act yet. For the first 15 minutes bottles traded for almost nothing, never above 130. Then the price shot up and first reached the cap at minute 19.
My trades. I sold my 60 bottles in three steps: 20 at 45 and 20 at 48 around minute 12, then the last 20 at 8,000 at minute 18, about a minute before the price first reached 25,000. Short selling was not allowed, so I couldn't take advantage of the ridiculously high prices. The software allowed it anyway, which I take to be a bug: 13 of the 40 seats sold more bottles than any seat ever held, and the course's own results show a seat ending the day at −5 bottles and another at −289,825K cash.
Result: 14th of about 40, finished flat, cash 161,859.60, or 2,698 a bottle, while bottles traded at the 25,000 cap for about 18 minutes. Not a good score. Waiting longer before selling would have let me sell into that peak.
Data: data/session-1
25 August 2026 · mtrader.org · 60 minutes
I did not take part in this session.
Setup. One stock in a limit order book, with brokers and news traders. News traders received private signals at random times, equal to the fundamental value plus up to ±4% noise, and were scored on P&L marked to fundamental value within a ±20,000 position limit.
Best bid and ask against fundamental value (log scale)
What happened. The market ran a bubble for the whole hour. 95% of the 4,279 trades went through at more than 2x the fundamental value, and the median trade was 6.7x the fundamental value. Buyers almost never found a price near value: the best ask sat above 2x value for 99% of the session. Sellers had the opposite luck: the best bid sat above 2x the fundamental value for 91% of it. For a news trader marked at value, every share sold above value was profit, and the bid offered that nearly all hour.
Data: data/session-2
7 September 2026 · mtrader.org · 60 minutes
Setup. Two stocks in limit order books, with three roles: asset exchangers, news traders and arbitrage traders. I was one of the nine news traders following STOCK2, scored on P&L marked to fundamental value, with a ±20,000 limit on each stock.
Best bid and ask against fundamental value
What happened. After a noisy first ten minutes, both books followed fundamental value for the rest of the hour: 70% of trades went through within 10% of value. No other session came close (why).
My trades. My best trade in terms of pure P&L was on STOCK1, where I had no signal: at minute 48 I sold 2,000 at 305 while it was worth 254, and bought them back at 240 seven minutes later. On STOCK2, the stock I received news on, I sold 2,000 at 117 at minute 24 while the news said it was worth 87, but I then stayed short as value climbed to 137 and had to buy 4,021 shares back at 140 to 148 at the close, which cost more than the sale had earned.
Result: 7th of the 9 STOCK2 news traders, P&L −39,633, both positions flat at the bell. STOCK1 earned 95,005 and STOCK2 lost 134,638. News traders were z-scored against the other news traders on their stock, on three things: P&L with final positions marked to fundamental value, breaches of the ±20,000 risk limit, and the ±20,000 target at the close, each stock separately. I breached neither, so the whole score came from the P&L.
Data: data/session-3
8 September 2026 · trading floor · 30 minutes
Setup. The session I enjoyed the most. The class became a trading floor, with no screen and no order book. Each of us was a broker with a number on a tag: odd numbers had to sell 30,000 shares while even numbers had to buy 30,000. Two brokers agreed a price for a quantity by voice and both wrote the trade on a paper card.
I was broker 10, a buyer. My card started at −30,000 shares and I brought it to zero in six trades with four other brokers.
Every trade of the session, with my six buys
- 17:16, 5,000 at 75 from broker 7. One of the first deals in the room, with no price yet to compare it with. The screen sessions had traded far higher, around 290 and 120 in Session 3, so 75 looked cheap. Bad idea: only one trade all session, at 99, went through at a higher price. It taught me to be patient, so that I am not the one who pays for price discovery.
- 17:41, 9,000 at 55 from broker 23. The session was nearly over and I still had a lot to buy, so I rushed and paid more than on the three trades before, but still about the volume-weighted price of that minute, so not a bad deal considering the quantity.
- Overall class trades: the prices paid slowly converged on the market's volume-weighted average price (market VWAP) for the session.
Result: task completed, 30,000 bought for 1,695,000, an average of 56.50. That was 2.14 above the session's volume-weighted price of 54.36, and ranked me 9th out of 13. Without the first trade my average would have been 52.80, which would have been 4th.
Average buy price of each of the 13 buyers
With nothing printed yet, I judged my first price against markets of previous sessions where prices stayed mostly above the 100 mark, and it cost me the most. Leaving 9,000 for the end cost me too. Spreading the buying across the half hour would have helped at both ends.
My trading card
14 September 2026 · mtrader.org · 60 minutes
Setup. A quote-driven market in two stocks. Dealers posted quotes, and everyone else could only take them with market orders. I did that as an asset exchanger with two jobs: sell 10,000 STOCK1 and buy 10,000 STOCK2 within the hour. I also had to stay within ±5,000 shares of a straight line from my starting position to zero.
What happened. As shown in the graph below, dealers often quoted far from value and wide, so crossing the spread was usually too expensive and I traded rarely: 43 fills, and only three of the 46 traders had fewer.
My orders against the traded price and the quoted spread
Each dot is one of my market orders, sized by the number of shares. A triangle marks an order whose price or spread was beyond the axis.
What went well
- Reacting quickly to the fat-finger mistake. At minute 50 a dealer bid 3,000 for STOCK2 (instead of 300, I suppose), and I filled the order right away. Buying the 1,000 shares back near 300 added about 2.7 million: without it I would have finished about 2.4 million down instead of 341,000 up.
- Selling STOCK1 early and relatively high with small spreads
- Buying at times of low spread on STOCK2
What went wrong
- Buying 1,000 STOCK1 at 285. At minute 45 I mixed up STOCK1 and STOCK2 and so whether to look at the bid or the ask. Selling them at 60 three minutes later cost 225,000.
- Waiting too long. The last 6,000 STOCK1 were sold at around 50, near the day's lows.
My position against the ±5,000 band
STOCK1 stayed inside the band all session. STOCK2 fell behind the line from about minute 33.
What the band was teaching. I guess it was the teacher's way of saying: buy a little all the time instead of a lot at the end. On STOCK2 I did the opposite and over the session I paid 297 a share, while the median STOCK2 trade went through at 217. Following the band with one simple rule, 1,000 shares every six minutes but waiting whenever the spread was above 20%, would have cost about 212 a share. The last 1,000 still had to be bought at a wider spread, because it never came back under 20% before the bell. Without that rule, one slot lands on an ask of 1,000 and the average climbs to 291, so steady buying only pays if you also refuse wide spreads. The one thing I would have given up is the fat-finger sale.
Data: data/session-4
Sessions 2, 3 and 4 all had a fundamental value that only some traders saw. Yet, the same class produced three different markets.
| Session 2 | Session 3 | Session 4 | |
|---|---|---|---|
| Trades within 10% of fv | 3% | 70% | 14% |
| Trades above 10x fv | 34% | 0% | 14% |
| Participants that received fv information | 24 of 41 | 20 of 40 | 14 of 56 |
| Who could post prices | everyone | everyone | the 28 dealers only |
| Opening | continuous, empty book | call auction (STOCK1) | continuous, empty books |
Below are my hypotheses to explain why prices followed fv or not, built from the course material and the data.
A. The opening set the tone. Sessions 2 and 4 opened into continuous trading from an empty book. Look at Session 2: its first trade went through at 110, and at second 20 one printed at 100,000, about 1,000x fv. The median trade in the first 5 minutes was 29x fv, and the market never came back: 34% of all its trades were >10x fv.
Now, what about Session 3? STOCK1 opened with a 5-minute call auction. It cleared at 235 while fv was 261, and then traded close to value from then on. The auction also gave informed traders time to act on what they knew: the first value push went out at second 0 and 11 more had arrived by the time the auction cleared at second 348, so the nine news traders on STOCK1 could price the auction from their signals. STOCK2 had no auction and opened at 994, 7x fv, but its median trade was within 20% of value from minute 10, possibly guided by STOCK1's price traders saw on the side of their screen.
B. More informed traders did not help. Agreement among them mattered more.
Session 2 had the larger informed group, 24 of its 41 seats, and because it traded a single stock all 24 were estimating the same value. Yet each of them received one of three noisy streams: each stream was off by about 1.3% on average and never more than 4%, with no bias up or down. The three streams were independent, so two traders could be up to 8% apart.
Session 3 split a smaller group across the 2 stocks, 9 news traders each, and gave everyone following a stock the same number: in the data the news matches the true value almost exactly.
Hence, traders working from the same number could pull a book towards it, while 24 working from three numbers 8% apart left enough disagreement for a bubble to sit in.
C. Two traders knew something nobody else did. Session 3's other informed seats were the 2 arbitrage traders (only there in Session 3), and what they were given was not either stock's value but the ratio between the two. That is a signal you can only trade by comparing the books, which is one more force pushing both prices towards the same relationship as the values.
D. Informed traders could post prices in Sessions 2 and 3, but not in Session 4. In Sessions 2 and 3, a trader with a value signal could post a limit order at that value. In Session 4, informed traders could only use market orders, and only dealers could quote. The dealer lecture from class explains it: dealers facing better-informed traders widen their spreads to cover what they lose to them, and move their quotes only after informed orders hit them. Session 4 also had the smallest informed share of the three, one trader in four.
E. The class had learnt from Session 2. It was the first time the class traded an order book against a fv, and the only practice before it was Session 1, where prices ran up to the cap. So the class came to Session 2 with no experience of trading around a value. The Session 2 results came out before Session 3, so by then the whole class had seen the scores from a market that ignored value for an hour.
F. Session 4 forced unreasonable quoting. Dealers were required to make markets, and quotes as high as 10,000 appeared in both stocks. Such quotes keep a dealer present without trading much, but a market order that reaches one still fills there, and 14% of Session 4 trades went through above 10x fv.
Conclusion: only Session 3 had all six conditions: an auction for the first STOCK1 price (A), one shared estimate per stock (B), two arbitrage traders tying the stocks together (C), informed traders who could post prices (D), a class that had already seen what a bubble costs (E), and no one forced to quote prices they did not mean (F).
The class traded through a browser screen that redrew the tape only now and then and kept no history. I wanted every trade and every book change as it happened, in a second window, without touching the tab I was trading in. So I built it using Claude Code.
trading tab on Chrome → Chrome extension → local server (saves every message) → viewer
A Chrome extension replaces the page's WebSocket constructor before the page loads, so every connection the page opens gets a listener, reconnects included. The extension forwards the messages to a small server on my machine, which saves them and feeds the viewer window. FTS Web Trader (Session 1) never reports your own fills, so the server infers them from changes in cash and holdings; mtrader.org (Sessions 2 to 4) acknowledges every fill, so nothing there has to be inferred. Recording was read-only; the tool never sent an order.
data/ holds the anonymised market data, where every trader is a random label.
charts/ holds the ten charts: eight drawn in R from that data, and the two open-outcry
charts, built from the course's trade record, which is not published. images/ holds
my open-outcry trading card.
Built with Claude Code.