This thesis is one of the first research studies about weather derivatives. Firstly, it focuses on the modelling of temperature, which is the most frequent weather variable (chapter 1). An autoregressive univariate model with periodic volatility is shown to be appropriate to describe its dynamics, especially with French data. Then, it addresses some financial aspects of weather risks. In particular, the fact that market and temperature are quasi independent seems to justify the common use of an actuarial framework for valuation (chapter 2). Finally, the model risk corresponding to this practise is quantified in chapter 3; whereas Futures prices appear to be robust to modelling errors, large uncertainties are shown around option prices. This source of uncertainty has been identified: it concerns trend and seasonality, the deterministic components of the temperature process mean.
