How the diversification check is measured
The free check turns a list of holdings into one figure: the number of independent bets it represents. Here is exactly how, and where it stops.
1. The data
Each ticker is matched to three years of weekly closing prices (the last close of each week), adjusted for splits and, where the source provides it, for dividends (the data file states the source of each series). The reference list is the universe followed by Helios, the system of Bayes Solutions: about 2,900 stocks, ETFs and funds listed in the United States, Europe and Asia, plus the US index funds most commonly held by US investors. A ticker needs three full years of history to be listed. The window and the list are shown on the check page under the tool. The data is refreshed every week; it is descriptive market data and nothing else.
2. Correlation of weekly moves
For every pair of lines we compute the correlation of their weekly moves over the same 156 weeks. Weekly moves are used because markets in Asia, Europe and the United States do not close at the same hour: day by day, the same move would be counted on different days. Your browser rebuilds each correlation from a compact file, to within 0.005. A value of 1 means the two lines moved together every week, 0 means no link, negative values mean opposite moves. This is the matrix shown on the page.
3. Observed volatility
For each line we compute the annualized volatility of its weekly moves over the same window. Volatility is what turns a weight into a share of risk: a bond fund at 40% of the weights usually carries far less than 40% of the risk.
4. The risk of the whole, and its independent components
Your weights are scaled to 100% and combined with the volatilities and correlations into the variance of the whole portfolio. That variance is then split into independent components (the principal components of the covariance of your lines). Each component carries a share of the total risk; those shares add up to 100%.
5. The number of independent bets
The figure is the exponential of the entropy of those shares. If one component carries all the risk, the entropy is zero and the figure is 1: whatever the number of lines, there is one bet. If eight components each carry one eighth of the risk, the figure is 8. Eight lines that move together land near 1; eight lines that never move together land near 8. The page also shows each line's share of weights and of risk, and the largest bucket and region by weight.
Where it stops
- Three years is a short window. Change the period and the figure changes. We state the window rather than hide it.
- Correlations rise when markets fall. The figure is therefore on the high side for stressed periods.
- No fees, taxes or cash flows are deducted. Lines outside the reference list are left out, and the page says which.
- The method describes what a portfolio contains. It does not tell anyone what to hold, and it takes no account of anyone's own situation.