Day-to-day risk can be measured in Matrics by an adapted version of Value-at-Risk (VaR). The VaR is a single number giving the maximum likely loss over a specified time horizon, e.g. one day or one week, and is the standard way to measure short-term risk in liquid trading portfolios.

The advantage of VaR is that it is easy to communicate and
suitable for setting trading limits, e.g. VaR should not exceed
*x*% of the trading capital. It is also a fairly
automatic method in that most model parameters can be estimated
from historical data. Note that VaR is only
relevant for actively traded portfolios that may change between
short and long positions in few days. For evaluation of
long-term risk in more static portfolios, we recommend
scenario analysis instead.

Once the VaR model is configured, VaR numbers are immediately available in the same way as positions and P/L in any standard Matrics report. The VaR model in Matrics is completely analytic and sufficiently fast to allow instant on-line calculation, even with any general break-down in the report. Note that VaR numbers are not additive. For example, the total VaR of all portfolios in an organization will normally be smaller the sum of the individual VaRs for each portfolio. Unless the portfolios are perfectly correlated, this is a natural consequence of risk diversification between the portfolios.

Making a good implementation of VaR for the freight market is difficult for several reasons. The model must account correctly for all correlations in the time dimension as well as in the geographical dimension. The model should, for example, see that the risk of spreading the two nearest quarters is much higher than the risk from spreading two quarters far out on the curves. (This is because the far end of the curve tends to move in parallel shifts while the correlations are much lower in the near end.) Likewise, the model must be able to see the risk involved in any kind of segment spread or geographical spread, e.g. Supramax vs. Panamax, P1A vs. PM4TC, etc. The standard approach is to maintain a large covariance matrix for each combination of all traded products. Unfortunately, this works very poorly for freight portfolios with physical exposures because the dimensions of the covariance matrix get impractically large (there are a large number of single route indices and the time dimensions is essentially infinite). Furthermore, there are no forward prices available for estimation of most of the single routes.

Matrics solves this in a number of innovative ways. Volatility and correlation for the liquid indices are estimated directly from market prices. The less liquid indices are linked to the more liquid ones via a model that includes estimates of the spread risk. The price movements in the freight market also have heavier tails than captured by the normal (Gaussian) distribution used in most analytic VaR models. (Heavier tails means that a larger proportion of the days have small price changes and very large price changes than what is predicted by the normal model.) This is countered in many systems by using simulation based VaR models. These can easily work with heavy-tailed distributions, but are unfortunately much slower than their analytic counterparts. Matrics combines the best of both worlds by using a custom analytic model which is very fast and can also incorporate any realistic degree of heavy tails.