This entry pertains to the 3rd Annual Valuation Workshop held at The Wharton School on June 13-14, 2024. I summarize my reactions to each of the papers followed by some outstanding questions at the end.

Introduction

The conference was primarily about valuations. Valuation ratios (of various kinds) were featured quite prominently, and the two main questions of interest were (i) how do we explain movements in valuations, and (ii) how are valuations today useful for forecasting future movements in the stock market? Each paper touched on either (i) or (ii) (or even both!).

Day 1

• Ben Knox talked about how to decompose each period’s returns into components. This echoes some of the themes in Richard Roll’s famous 1988 AFA address titled “R2” where he argues that even with hindsight, the ability to explain stock price changes is modest. In the authors’ approach, there are some key assumptions being made. The first assumption disciplines how the expected return on the market is connected to the expected return on dividend strips, and the other is a Taylor approximation which generates the linear decomposition (a standard asset pricing trick).
• Ricardo De la O provided a cross-sectional decomposition using professional forecasts, which is a follow-up to their previous work. They rationalize their empirical pattern with a model, which basically says that constant gain learning can generate the patterns in survey data. It’s a reduced-form model in the sense that constant gain is imposed (as opposed to being micro-founded through mechanisms like fading memory). It would be interesting to see what implications a more standard Bayesian learning model would have, which is prevalent in the literature.
• Paul Decaire presented his work in which he collected data from equity reports, enabling them to observe the entire discounted cash flow (DCF) model. The “price” measure is the 12-month target price that analysts derive, and they provide a granular decomposition of what drives the variation in the inputs into the final price estimate. In its current form, the paper provides very good insights into analyst decision-making. What seemed incomplete in the current version is the link to actual market prices, which the authors say is still in progress. One result they had related to this is that analysts’ security market line is steeper than the econometrician’s, which suggests that analysts may be overestimating the risk premium that investors demand for holding risky securities.
• Sebastian Hillenbrand talked about the construction of the “optimal” stock-valuation ratio. The main conclusion is that we should be using forward-looking price ratios rather than trailing price ratios, as trailing price ratios also contain expected cash flow growth. They show that the outperformance in forecasting ability is real, both in-sample and out-of-sample. This result is interesting because we typically think there is a trade-off between maximizing in-sample \(R^2\) and out-of-sample \(R^2\).
• Toomas Laarits presented his finding on how the risk-return tradeoff of timing strategies deteriorates substantially as the investment horizon increases. He then provides a model with seasonalities in the volatility of the state variables, which makes timing strategies profitable in the short-run but not so much in the long-run.

Day 2

• Andrei Goncalves presented his new work (he said he worked until 3am the morning before the presentation!) on the subjective expected returns of institutional investors. He observes expected return, subjective volatility, and correlation across asset classes, which enables him to ask which components drive variation in expected returns: risk price vs. risk quantities. It would be useful to see if the predictions on holdings (quantities) given by these results also hold up in the data.
• Kristy Jansen presented new work on treasury demand. The authors have undertaken a gigantic data exercise to construct the most comprehensive dataset on treasury holdings of investors (from call reports, ETF Global, TIC, CFTC, eMAXX, IMoneyNet, Morningstar, Federal Reserve). They then estimate demand curves of each investor and embed these estimates in a Treasury equilibrium model. There were a lot of discussions about the validity of the instrument (they use monetary policy shocks), primarily the fact that changes in the level of rates can affect demand through channels other than prices (e.g., by relaxing financial constraints or strengthening the reach-for-yield incentives).
• Zhengyang Jiang presented a theoretical paper on how investor composition matters for the effectiveness of quantitative easing (QE) and quantitative tightening (QT). The main idea is that when the central bank implements QT followed by QE, the bond investors are starting from a lower bond position below their steady state. This is because in the model, bond investors are assumed to have portfolio adjustment costs. Since the liquidity traders have a lower risk-taking capacity, they demand a higher expected return to hold more bonds, leading to a lower bond price below the pre-shock level. I liked the model’s simple mechanism and its emphasis on investor composition, which made a lot of sense.
• Lukas Kremens presented his paper that shows how subjective exchange rate expectations can forecast currency appreciation at the two-year horizon, both in and out of sample. This is a win in the international finance literature where pretty much “nothing predicts foreign exchange rates.” They show that these expectations can be explained by economic fundamentals and that after controlling for the three macro-finance variables, the residual information in survey expectations does not forecast currency appreciation in their sample. It’s a punchy paper with easy-to-digest findings and important implications.
• Thummim Cho talked about his paper where the main object of interest is the percentage gap between price and the present value of dividends computed with a chosen asset pricing model (i.e., a particular stochastic discount factor (SDF)). The key insight is that today’s abnormal price is the expectation of a simple discounted sum of conditional abnormal returns. Based on this identity, the paper develops a calendar-time estimator for abnormal price that resembles the time-series regression estimator for abnormal returns. If I understand correctly, this is how it works. Suppose we have a stock that is currently trading at $100 per share. We want to determine if this price is abnormally high or low compared to the present value of its future dividends, as implied by the CAPM model.
  To do this, we first need forecasts of the stock’s future annual abnormal returns (alphas) over some long horizon, say 10 years. Let’s assume these forecasted annual CAPM alphas are: 5%, 4%, 3%, 2%, 1%, 0%, -1%, -2%, -3%, -4%. Next, we discount each of these future alphas back to the present using the CAPM’s stochastic discount factor (SDF). This SDF accounts for both time-value of money and risk adjustments. If this discounted sum is negative, it means the current $100 price is higher than the CAPM’s present value of dividends - i.e., the stock is overpriced. Conversely, a positive sum implies underpricing. Obviously, the estimation errors for the SDF and the alphas propagate throughout the process, but I thought the main insight — that mispricing can be rewritten as a discounted sum of conditional abnormal returns — was cool.

Takeaways

1. To understand stock price (\(P\)) dynamics, we often use valuation ratios (\(P/E\), \(P/B\), etc.) and study their covariance with another object. Often, the denominator ($E$) is more volatile than \(P\), so the dynamics of the valuation ratios may not necessarily reflect the dynamics of \(P\). This is a point that repeatedly came up at the conference, and smoothing the denominator is one obvious solution to make the series less volatile. However, what if the unsmoothed valuation ratio is really the object of interest?
2. A great conference needs to have (i) “discussable” papers, (ii) a great audience, and (iii) great food. This conference had all three, and I’m very happy that I was able to attend the workshop this year.
3. Collecting new data sources or constructing novel datasets is crucial for empirical research, as showcased by several papers at the conference that leveraged unique data to provide fresh insights.