Seeing where the wind blows me... 🌻

• Data Engineer at Swiss Re
• Summer Intern at Swiss Life Asset Managers
• Spring Insight at Barclays

• svi-py Building volatility surfaces using the stochastic-volatility inspired family of IV parametrisations.
• pymgarch Multivariate GARCH for Python by porting tsmgarch (R package) to Python and extending it.
• polygon-options-puller Pull, reformat, and store Massive/Polygon S3 flat files.
• interest-rate-models Access different interest rate models: short-rate models (Vasicek, CIR, Ho-Lee, Hull-White, G2++) and full-curve models (Heath-Jarrow-Morton, LIBOR Market Model) behind one interface

Title: Mean Reversion in the Intraday Implied Volatility Surface of S&P 500 Options

We study intraday S&P 500 index option implied volatilities using an extended stochastic volatility-inspired parametrisation, recalibrated jointly across all expiries at 60-second inter vals over 63 trading days. Two experiments are conducted. The first decomposes the recon structed implied volatility surface via functional principal component analysis, fits a vector autoregression to the resulting factor scores, and tests whether the surface-level forecast can outperform a random walk. The second measures the basis between each quoted implied volatility and the fitted surface, tests for serial dependence, and assesses the economic scale of the resulting deviations against the bid–ask spread.

Title: Reverse-Engineering a Dominant Market Maker from Level 4 Order Book Data (ongoing)

Avellaneda and Stoikov's definition of market making as a stochastic optimal control problem is one of the seminal works in the academic study of market making. However, testing whether a large liquidity provider actually follows this method requires detailed order book data, not commonly disseminated by traditional financial exchanges. Recently, a Level 4 order book dataset from the perpetual futures exchange Hyperliquid was published, for the first time allowing us to pose the question whether any large market makers actually use this model. We identify the largest market maker by traded- against resting notional, reconstructing its quotes, quoting ladder, and inventory every second. Arrival intensities are estimated from the wallet's own resting orders by censored Poisson maximum likelihood estimation. We find the wallet's quoting mostly inconsistent with the model tested.