The most recent Barra US Equity Model, USE4, contains some important innovations in factor risk modelling, including the introduction of country risk factors, volatility regime adjustments, and eigenfactor risk adjustments. Amanda White spoke to executive director and head of equity factor model research at MSCI, Jose Menchero, about what that means.
Historically there have been a number of shortcomings in the way risk models traditionally have been calculated. Generally, risk has been underestimated in optimised portfolios, biases have not been encapsulated in modelling, and the fact that volatilities are not stable over time has not been adequately addressed.
Following many years of research, the new release of the Barra US Equity Model overcomes many of those shortcomings by introducing several important innovations.
One of the advances in this risk model is the introduction of a country factor. Until now, Barra single-country risk models have not included a country factor, just industry and style factors.
By introducing the country factor, executive director and head of equity factor model research at MSCI, Jose Menchero (pictured) says, the interpretation of the other factors changes.
“The factors represent the underlying drivers of equity returns,” he says. “By including the country factor, the industry factors are reinterpreted to represent the industry net of the overall market.”
Essentially it becomes the performance of the industry relative to the market and captures the intuitive interpretation of the industry.
But this country factor is not just an interpretation issue; it has an important effect on the risk forecast accuracy of the portfolio, because it disentangles the market effect from the industry effect.
Since industries have a common exposure to the country factor, it captures and quantifies the empirical observation that industries become more correlated in times of financial crisis.
MSCI will adapt this same methodology to other single-country risk models, including Canada, Australia and China, and will eventually roll out models for other major markets.
The US country risk model contains 60 industry factors, 12 style factors, and one country factor. In smaller markets the expectation is there will be fewer industry factors. For example, in Australia or Canada there might be 15 to 20 industry factors, but the basic structure of the model will be the same.
It is not just the introduction of the country risk factor that is innovative in this model. Menchero says there is a whole set of innovations in the USE4 model that will be taken to the other single-country models.
Two other major innovations in the model are the eigenfactor risk adjustment, which reduces the effects of sampling error on the factor covariance matrix, and a volatility regime adjustment, which is designed to calibrate factor volatilities and specific risk forecasts to current market levels.
“One of the long-standing problems in risk models is a tendency to under-predict the risk of optimised portfolios,” says Menchero.
“We have conducted research into this topic and found the root cause to be estimation error due to noise and finite sampling.”
MSCI has found a way to make adjustments to the covariance matrix to reduce this bias, which it calls the eigenfactor risk adjustment.
“We estimate the covariance matrix then identify special portfolios that capture the systematic bias,” says Menchero, who is a former professor of physics at the University of Rio de Janeiro in Brazil.
“Each eigenfactor is a portfolio of the original factors, such as banking industry or value style, and are characterised by not being correlated with each other – they’re called eigenfactors. The predicted volatilities of the eigenfactors are systematically biased and we can adjust the covariance matrix to remove those biases.”
Another innovation, the volatility regime adjustment, addresses the challenge of predicting the volatility of a portfolio and, in particular, the fact it is not stable over time.
“The challenge of building a risk model is that we have no crystal ball and must rely on history,” says Menchero.
“So going into a crisis, how can we use history to better predict current volatility?”
This new method looks across all factors on a given day, analyses the returns of those factors compared to their volatilities and can reduce the effect of those biases by calibrating factors to given market conditions.
The effect is it reduces the tendency to under-predict risk going into a crisis, and more accurately uses historical data to predict current volatility levels.
The new model, which was launched during the huge market meltdown in August – and so providing almost free PR – has been presented to about 100 clients, hedge funds and funds managers in the past few months.
MSCI will continue to provide USE3 (the legacy Barra model), as it is embedded in the investment process of many fund managers, and there is recognition it’s not easy to switch quickly. But Menchero says the new model represents a significant advance over the USE3 model.
