Dynamic MaxVaR

In 2006-07 I worked on a model to dynamically predict MaxVaR, a measure for portfolio risk during a time period (and not at the end of it, which is the plain old Value at Risk). MaxVaR is the lowest return on your portfolio, with a specified confidence level, during the next N days. Of course it is lower than VaR (both VaR and MaxVaR are negative for reasonable horizons like 10-days). Key features of the model are:

1. ARMA-GARCH model for conditional mean and variance

2. Pearson Type IV distribution to model negativeley skewed and leptokurtic (standardised) residuals - both MLE and Method of Moments Estimation

3. 10,000 Monte Carlo Simulations to compute MaxVaR

4. Testing on 20 years' data from FTSE 100, Dow Jones, CAC 40, Australian All Ordinaries and Swiss Market Index

You can download the software (MATLAB and C code, released under GNU General Public License) here. For an old manuscript of the work, click here (please email for the published version). The work was also invited for presentation at the 12th International Conference Conference on Applied Stochastic Models and Data Analysis (http://www.asmda.com/id7.html), held in Greece (May-Jun 2007). 

After two rounds of review and two years of long wait after the initial submission, this work was published in the peer reviewed journal Quantitative Finance. The citation is

Bhattacharyya, Malay, Misra, Nityanand and Kodase, Bharat (December 2009). MaxVaR for non-normal and heteroskedastic returns, Quantitative Finance, Vol. 9, No. 8, 925 — 935.

A future track could be to use the Johnson SU distribution with a GARCH model to address the conditional asymmetry and leptokurtosis of returns.

Plain Old Value at Risk

Here is some MATLAB code (again under GNU GPL) to compute the plain old Value at Risk  using three different methods - plain old static VaR, Riskmetrics and Extreme Value Theory (both Block Maxima and the Point Over Threshold method).  Have included a sample database, help file coming soon. Check out the report here.

Guest Lecture at IIM Bangalore

In November 2008, I delivered a guest lecture to second year MBA students at my alma mater, the Indian Institute of Management Bangalore (IIMB). This was for the elective Financial Time Series Analysis offered by Prof. Malay Bhattacharyya. You can download the lecture notes here. Please discount the horrible look and feel of the presentation as that is the best I could do with Microsoft Powerpoint.

Workshop at IIT Kanpur

In March 2009, I conducted a 2-day workshop for students at IIT Kanpur with two colleagues at Goldman Sachs on Financial Mathematics and Careers in Finance. You can download the slides used for one of the presentations, Statistical, Econometric and Time Series Modeling in Finance, here. Please do pat me on the back for the wonderful look-and-feel of the presentation, this time wisdom dawned and I used Latex.

Research fields I want to explore

1. Markov Chain Monte Carlo

2. Bayesian Inference Models

3. Copula Modelling 

4. Lasso (LARS) and Dantzig Selection

For help or suggestions, you are welcome to mail me at nmisra [AT] gmail [COM].