Every number from assembling the machine to the input-sensitivity failure is explicitly synthetic, chosen to teach the mechanics honestly. Building a real CMA table means pulling defensible inputs asset class by asset class, and knowing which leg is hardest. The equity expected return reuses machinery you already own in full: M3.02's historical-versus-implied equity risk premium toolkit, added to a currency-matched, cleaned risk-free rate (M3.02's country risk premium page and M3.03), is exactly the equity μ input this node needs, there is nothing new to learn there, only a pointer back. The bond leg is comparatively the easiest: a bond index's current yield to maturity is a direct, observable, forward-looking estimate of its expected return, unlike equities where the premium must be extracted rather than read off a screen. Gold has no cash flow and no dividend discount model to invert; treat any assumed real return for it as a contested, thinly evidenced number and hold it loosely. REITs can be estimated income-statement-style, current distribution yield plus expected NOI (net operating income) growth, echoing a cap-rate logic, but a nascent market (India's listed REIT sector dates only to 2019) carries extra estimation risk from short history and thin trading. Alts deserve their own warning, flagged already in assembling the machine: reported correlations and volatilities for private equity, hedge funds, and other appraisal-priced sleeves are systematically understated, because infrequent, model-based marks smooth away the true, economic volatility and correlation to public markets, an institution that takes a factsheet's "low correlation to equities" at face value is very often buying a statistical illusion, not a real diversifier (Common Mistakes, item 8).
AA1.04 is where this section's subject actually lives. What follows here is the short form, kept because a frontier is unreadable without knowing where its inputs came from. AA1.04 owns the full treatment: the framework for setting capital-market expectations, the specific biases that corrupt them, the business-cycle and Taylor-rule anchors for the cash and bond legs, the Grinold-Kroner decomposition for equities, and the smoothing correction that makes an appraisal-priced sleeve's risk statistics usable. Read it before building a real table, and treat this section as the pointer it is.
Advertisement
Where can you get capital market assumptions?
For real, current inputs: J.P. Morgan's, BlackRock's, and Vanguard's annually published long-term capital-market-assumptions sets are free, credible, and give you a full cross-asset table built by professional teams, the natural starting point for the equity, bond, and alternative-asset legs alike. Damodaran's implied-ERP and country-risk pages (already your source since the live-numbers ritual in M3.02) remain the right place to build the equity leg yourself rather than take anyone else's number uncritically. Whatever the source, the rule from assembling the machine does not relax: run the PSD check on the resulting correlation matrix before a single frontier number is computed from it.