APPENDING SOUL TO OUR Models

On 15.01.2015, Swiss National Bank (SNB) announced it was to impose a negative interest rate of -0.75% on deposit from 22. January. 2015. The sudden announcement shocked the market. Any quantitative interest rate or FX model simply based on recent historical data would fail.

READ MORE

TOO MANY MODELS?

In option valuation, the price is predominantly a function of volatility value. Based on different assumptions, there exists various volatility models. To decide which one to use is not as easy as it sounds. An online dynamic model selection framework should be introduced into our daily work.

READ MORE

BIG DATA, MACHINE LEARNING AND FINANCE

Big Data has been such a popular term. But what is Big Data exactly? What does it bring us in our daily  modeling work? To efficiently analysis and process large quantities of data in the new age, many exceptional technologies have been applied. Machine Learning is a power tool to extract information or pattern within the big data set. There are several Machine Learning approaches ...

Read more

Quantitatively construct Portfolio

Designing the correct portfolio cannot be done by human intuition alone. We often hear investors claims 'We don’t know at what proportion we should hold those assets, so we invest on them equally.'. Facing the complex financial market consisting of thousands of assets, even the most experienced investors would have difficulties to make decision simply based on intuition. It requires help from sophisticated mathematical tools.

READ MORE

MI PRODUCTS

Gaussian Process and supervised Learning

Gaussian Process regression is a non-parametric Bayesian approach. The term non-parametric suggests it lets the data 'speak' more for the shape of the model rather than a pre-human-assumed model shape.

READ MORE

WHEN BELOVED normal DISTRIBUTION FAILS

Due to its nice properties, normal distribution is widely used in different fields. E.g. David Li’s model uses Gaussian copula to simulate the joint default distribution to price CDS, which is arguably blamed as “recipe for disaster” of crisis 2008 - 2009. It is known that normal distribution is incapable of modeling heavily tailed problem. What are the alternatives?

READ MORE

MI Group