Nicolas Liebaert Français

Rebuilding the payoffs I priced on the desk

I did two internships on structured products desks. I then wrote my own code for the same payoffs, to see how each price is built. That code runs on this site, and each study says what it could not show.

Natixis CIB
Sales Assistant, equity structured products, France distribution, 2025
Kepler Cheuvreux
Structuring Assistant, cross-asset excluding FX, 2024
SKEMA Dubai
MSc Financial Markets & Investments
The 2s5s10s butterfly since 1990, drawn at averaging windows from 5 to 500 days.
8.6% of paths lose capital on a Phoenix autocall at the pricing drift. At a drift of 7.5% a year, it is 2.4%.
0.33 Sharpe ratio after costs on the 2s5s10s butterfly over 36 years. It falls to 0.17 if trades fill one day later.
16× less error from 101 Leisen-Reimer steps than from 1,010 Cox-Ross-Rubinstein steps. It ranges from 14× to 46× between 1,005 and 1,015 steps.

Pricing: my own code, after my internships

Studies of the payoffs and greeks I met in my internships, in code I wrote afterwards. The Python code runs in your browser.

These two pages download Python, NumPy and SciPy once, about 31 MB, from this site. They run in a background thread, so you can use the page while they start.

Source: structured-products-pricer, option-pricer

From my own research

Trading rules tested on historical market data, after costs.

Source: relative-value-strategies, position-sizing

How the site works

On the pricing pages, your browser recomputes the price with the same Python code as in my repositories. The sliders use 20,000 Monte Carlo paths. A high precision button runs 200,000 and reproduces the numbers in the READMEs.

The spread trade pages use JavaScript, which starts at once on a phone. An automated test runs the JavaScript and the Python on the same data and fails if any result differs. The data comes from FRED and the EIA and is refreshed every morning.

The pricing models use Black-Scholes with one flat volatility. This shows the shape of a payoff and its risk. A desk would not quote prices from it. Each page says where its model stops being reliable.