Activity: Talk or presentation › Talk at conference or symposium › Science to science
Functional data analysis is an ever-growing field in statistics where underlying data structures are believed to be functional rather than discrete. Of course, in real life all data is sampled discretely and one needs to take preprocessing steps in order to transform them into their supposedly underlying smooth functional form. The most commonly used methods of obtaining smooth functions employ fitting curves using B-spline and Fourier bases with a roughness penalty. How- ever, these methods tend to produce residuals that show unwanted correlation structures. On the hunt for a more canonical approach, we are applying the theory of factor models in order to obtain estimates for the underlying smooth function of the given data. By its very nature, our approach produces uncor- related residuals. We aim to investigate this idea further by using simulation studies and real life data. In this talk, we wish to present our findings and fail- ures in order to give a deeper insight into whether factor models and functional data analysis have more common ground than commonly assumed.