A Demand System Approach to Asset Pricing
Ralph S J Koijen, Motohiro Yogo
https://www.youtube.com/watch?v=LNwYHXdjoYc
Permanent Notes
Highlights
Think about asset pricing as a demand system. Movements in each assets can be traced back to specific investors.
Estimate demand curve for each investor. Demand for each investor is function of asset prices, asset characteristics, and demand shocks.
Heterogenous beliefs lead to different return expectation to the same asset characteristics:
Parsimonious representation of factor structure in returns consists of market equity (market cap), book equity, profitability, investment, and market beta:
Can think about polynomials as approximation to exponential curves:
Characteristics-based demand function:
Latent demand is demand for characteristics that are ubobserved:
Can think of weighted latent demand as measure of investor sentiment toward a stock and dispersion across investors as measure of disagreement:
Heterogeneous demand elasticities:
Median institution holds 67 stocks:
Market equity is endogenous, as in its almost certainly correlated with latent demand. Thus can’t simply run OLS on regression in market equity included as the error term / latent demand is correlated with one of the regressors (market equity). To get around this, authors instrument for market equity on a per investor basis using the AUM-weighted prevalence of a stock within the investable universe of OTHER investors, the idea being that this variation is exogenous relative to latent demand of a particular, held-out investors. It’s the market equity the stock would have if all the investors who could invest in it invested an equal weighted share in it (1/N rule, effectively). I’d imagine this would yield fairly similar market equity for a given stock across each investor, but I guess it’s important not to include the investors own ownership in the instrument in order not to pollute it:
Instrument works well because different institutions have small and sufficiently varied investment universes, so you actually get some variation across stocks. Wouldn’t work well if everyone was basically investing in the same set of companies:
Strong first stage, though I’m not sure what they mean when they say “minimum across institutions”:
Households are more elastic than mutual funds, banks, pension funds etc and have become more elastic over time (Robinhood?). Mutual fund are relatively less elastic.
Disagreement among households has increased over time, spiked in 2008:
Average price impacts of a hypothetical drawdown have lowed over time across institution types:
Variance decomposition of cross-sectional returns. Variance in returns across stocks largely driven by extensive margin of latent demand, as in portfolio weights among stocks that are held by investor. Thus returns are mostly explained by demand shocks that are unrelated to changes in observed characteristics in the stocks themselves, which is a big deal (The Inelastic Demand for Startup Equity):
Variance decomposition results show very clearly that individual investor sentiment (both its magnitude and dispersion) meaningfully influences returns:
Mean reversion in latent demand implies return predictability, since those stock with high latent demand will see their prices fall or rise more slowly as latent demand falls toward zero/one. Expected returns based on mean reversion over subsequent month appear to explain some portion of realized excess returns (and an even higher portion for small stocks):
Paper address point Nassim Taleb makes in Fooled by Randomness in reference to Robert Shiller that stock prices move too much relative to plausible changes in fundamental value, new news, etc. Cross-sectional stock returns are mostly explained by demand shocks that are unrelated to changes in stock characteristics themselves:
