 |
EM Applications risk models are derived from a proprietary application of the EM ("Expectations maximisation") algorithm.
The use of EM has multiple advantages:
- Stock returns are the only input data requirement
- International comparisons of accounting data are unnecessary
- Pre-determination of sector categorisation is not needed
- Observation frequency can be daily, weekly or monthly
- The model can be time- or stock-weighted
- Identifiable time-series may be pre-specified as factors
- Any stock characteristics may be mapped onto the factors
a. To test if the characteristic is associated with risk
b. To examine the characteristic’s contribution to portfolio risk
EMA has published proofs to demonstrate that, by comparison with “Fundamental”, “Macro-Economic” and Principal Components based “Statistical” models, EM is the least assumptive approach, the only one certain to converge to a solution.
EMA produces both standard models which have been tried and tested over many years, as well as custom models for specific applications. The standard models, covering all countries and regions around the world, are based on 200 weeks of unweighted returns data, generating 20-factor models. Experience shows that these standard models are extremely successful at forecasting meaningful 12 month risk estimates.
We believe we are the only commercially available solution for investment firms who operate on different time frames or would like to use proprietary modelling assumptions. EMA offers such clients the same institutional quality tools that we use to build our standard models.
Model Coverage:
Broad security coverage is guaranteed with statistical modelling techniques and EMA employs proprietary regression algorithms to ensure the most comprehensive coverage feasible. For example, EMA's standard equity risk product comes packaged with nearly 60 regional and country models encompassing over 60,000 unique equities, currencies and indices.
The mixed asset model can handle any asset type such as bonds, equities, FX and derivatives. The mixed model combines the best of both worlds - a true factor model coupled with sophisticated pricing models. This enables comprehensive yet intuitive risk attribution whilst at the same time properly modelling complex non-linear instruments given changing parameters like time, interest rates and instrument volatility.
|
|
