The HiddenLevers model is currently composed of two levels; a regression model that calculates the relationships between every economic lever and every asset, and an intelligent filtering process that separates out correlation from causation within this large universe of regression data.
The major components of HiddenLevers model are:
• Scenarios - A scenario is a representation of a major macro-economic or geopolitical event which has the potential to impact investment returns.
• Levers - HiddenLevers tracks over 130 different levers (economic indicators), including major macro statistics like GDP growth and CPI, market data like commodities and currency prices, and industry-specific levers like shipping rates and housing starts. HiddenLevers stores historical data for all levers, with up to 100 years of data available in some cases, and new levers continue to be added over time, increasing HiddenLevers' modeling capability.
• Industries - HiddenLevers has created a proprietary industry list, adding new industries where needed to capture the intricacies of the modern economy.
• Assets - The HiddenLevers model can be used with a wide range of asset classes, including stocks, ETFs, mutual funds, closed-end funds, options, and bonds. End-of-day pricing is used for all modeling and analytics, since macro-economic events tend unfold over the course of weeks, months, and years (though they can have short-term impacts). HiddenLevers maintains two decades of historical pricing data for its assets where possible.
The scenarios, levers, industries, and assets in the HiddenLevers model interact to provide projections of asset returns in a scenario:
Scenario → Levers → Assets → Portfolio Return
A scenario pushes levers up or down, which in turn push assets up or down, which in turn impact a portfolio's potential return in the scenario.
Margin of Error
The model is then run 2500 times for each scenario/portfolio combination. In each iteration, the model projects the returns for each asset using the historical regression coefficients for each lever, and using the scenario assumptions on how each lever will change. The model varies the regression coefficients for each iteration using a normal distribution around their mean (similar to a Monte Carlo model's varying of expected returns across iterations), and aggregates the results of the 2500 iterations to find a mean portfolio return with a 95% confidence interval. The confidence interval is displayed on the report as "margin of error" for each scenario.
What this means in practice is that for positions that are impacted by levers with a large standard deviation, the margin of error will be larger. For example:
Stock A has an Aluminum beta of 1.2. The Aluminum lever has a high standard deviation and the model uses a beta range of 1.0-1.4 around its mean to calculate the margin of error.
Stock B has a S&P beta of 1.2. The S&P lever has a lower standard deviation and the model uses a beta range of 1.1-1.3 around its mean to calculate the margin of error.
So, stock A will show a higher margin of error than stock B because the Aluminum lever has higher standard deviation.
Also, the margin of error will tend to lower with more positions in the portfolio because because these forces average out as you expand the set.