During my final university year, I had the opportunity to work in the Asset Management division of Caisse des Dépôts et Consignations, a French public financial institution. This was also my first real encounter with research. My actuarial thesis focused on the Black-Litterman asset-allocation model and, more importantly, on what happened when some of its convenient assumptions stopped being convenient. 

I spent days - and quite a few nights - wondering whether the usual Gaussian assumption for asset returns could be relaxed. This eventually led me into multivariate \(\alpha\]-stable distributions, heavier tails, alternative risk measures and the broader problem of modelling events that occur rarely but matter enormously when they do. 

Extreme events had entered the room, and apparently they were here to stay. 

Who I am

My name is Mathys Frans Carmel Baldacchino, I am 26, French, and a compulsive fishing-tackle collector.

I followed an academic path mostly rythmed by equations: first mathematics, then a master's degree in actuarial sciences - essentially applied mathematics for financial and insurance risks. Along the way, stochastic calculus, statistics and extreme value theory became some of my favourite playgrounds.

After graduating, I joined the Insurance-Linked Securities team at AXA Investment Managers as a quantitative analyst. Hurricanes, earthquakes and other natural catastrophes became part of my daily work. I was fascinated by their scale, rarity and unpredictability, but also by the slightly strange challenge of putting a financial price on something as physical as a hurricane.

It also taught me an important lesson: modelling risk is not only about understanding the phenomenon. At some point, somebody has to make a decision with the result. 

Wanting to understand the catastrophe side more deeply, I moved to Zurich and joined Bernina Re as a catastrophe modeller. There, I learned how catastrophe models are actually used, questioned and adjusted in practice: digging through documentation, comparing simulations with historical events, testing assumptions and investigating why a model behaved the way it did. I also worked directly with underwriting and reinsurance decisions, which taught me to explain complicated things without automatically making the explanation complicated. 

At some point, I realized that the part I enjoyed most was research itself: finding a question, disappearing into papers and documentation for a while, building something, discovering that it does not work, figuring out why, and starting again.

That is when I created Catmosphere. 

Catmosphere is less a company than a place where I can let scientific ideas accumulate. The common theme is fairly loose on purpose: mathematics, weather, natural hazards, Earth sciences, spatial statistics, and anything else that makes me curious enough to lose track of time.

One of the first projects came from Kopri, where I investigated the presence of stinging nettles in France. We combined field observations with meteorological, geological and topographical information, eventually building a spatial Bayesian model to identify other areas where the plant was likely to occur. The idea was simple: use statistics to reduce the amount of countryside that humans had to explore manually.

Then Siena Capital Group - through Caterina Technologies - invited me to join their ship as Director of Research & Development and founding partner.

The problem was very different: how could we price relatively illiquid securities, particularly catastrophe bonds, frequently and using mostly open tools? 

Answering that question progressively dragged me deeper into catastrophe science. We developed daily pricing methodologies using Bayesian conditioning, but doing so required understanding how catastrophe risk itself evolved through time. Eventually, I built an in-house tropical-cyclone hazard framework derived from the open-source STORM model and extended it using climatological information. 

What began as a financial-pricing problem therefore became a weather-modelling problem. 

I worked on synthetic cyclone catalogues, regional seasonality, ENSO conditioning and physical constraints based on the atmospheric environment. The objective was to preserve the flexibility of a statistical model while preventing it from happily generating storms that made little physical sense. The resulting research mixed historical cyclone observations, synthetic tracks, weather reanalyses and seasonal forecasts. 

That period also taught me a different side of research. I was working not only with researchers, but also developers, executives and business teams. A technically elegant idea is not particularly useful if nobody else can understand, implement or use it. 

Eventually, the company's financial situation deteriorated, and I returned to Catmosphere full-time. 

IrriGator combines historical weather, deterministic and ensemble forecasts, soil information and crop modelling to reconstruct the state of a field and search for irrigation schedules that preserve yield while reducing unnecessary water use. It currently brings together sources such as ERA5-Land, Météo-France AROME and ECMWF forecasts with AquaCrop simulations, with backtesting used to check what would have happened in previous seasons.

And that is more or less how Catmosphere works.

I rarely start with a particular technology. I usually start with something I do not understand, then try to find out whether mathematics, statistics, code and enough stubbornness can make it slightly less mysterious. 

Soon afterwards, a conversation with a farmer friend produced another question: could weather forecasts and crop models be used to decide not only whether to irrigate, but when and how much?

That question became IrriGator.