Fundamental Valuation

How do recent advances in Generative Artificial Intelligence (AI), as epitomized by ChatGPT, impact firm valuations? In their May 2023 paper entitled “Generative AI and Firm Values”, Andrea Eisfeldt, Gregor Schubert and Miao Ben Zhang quantify workforce exposures to AI for publicly traded U.S. companies and translate those exposures into firm valuation effects. Specifically, they:

1. Measure job task AI exposures by asking ChatGPT to assess whether each of 19,265 occupational tasks could be done by the current ChatGPT or an enhanced future ChatGPT.
2. Measure occupational AI exposures by aggregating task-level AI exposures to occupations per the O*NET database.
3. Measure firm AI exposures by mapping occupations to publicly-traded firms based on millions of public employee profiles such as those in LinkedIn using data from Revelio Labs. They validate this measure based on mentions of AI in 2023 company earnings announcement call transcripts.
4. Quantify effects of AI exposure on firm valuations by examining how stocks of firms with varying exposures to AI react to the release of ChatGPT on November 15, 2022.

Using occupational task descriptions, firm employee job descriptions and returns of associated stocks from ChatGPT release through March 31, 2023, they find that:

Keep Reading

A subscriber asked about boosting the performance of the Simple Asset Class ETF Value Strategy (SACEVS) and the Simple Asset Class ETF Momentum Strategy (SACEMS), and thereby the Combined Value-Momentum Strategy (SACEVS-SACEMS), by substituting ProShares Ultra S&P500 (SSO) for SPDR S&P 500 ETF Trust (SPY) in these strategies whenever:

1. SPY is above its 200-day simple moving average (SMA200); and,
2. The CBOE Volatility Index (VIX) SMA200 is below 18.

Substitution of SSO for SPY applies to portfolio holdings, but not SACEMS asset ranking calculations. To investigate, we test all versions of SACEVS, SACEMS and monthly rebalanced 50% SACEVS-50% SACEMS (50-50) combinations. We limit SPY SMA200 and VIX SMA200 conditions to month ends as signals for next-month actions (no intra-month changes). We consider baseline SACEVS and SACEMS (holding SPY as indicated) and versions of SACEVS and SACEMS that always hold SSO instead of SPY as benchmarks. We look at average gross monthly return, standard deviation of monthly returns, monthly gross reward/risk (average monthly return divided by standard deviation), gross compound annual growth rate (CAGR), maximum drawdown (MaxDD) and gross annual Sharpe ratio as key performance metrics. In Sharpe ratio calculations, we employ the average monthly yield on 3-month U.S. Treasury bills during a year as the risk-free rate for that year. Using daily unadjusted SPY and VIX values for SMA200 calculations since early September 2005 and monthly total returns for SSO since inception in June 2006 to modify SACEVS and SACEMS inputs, all through February 2023, we find that: Keep Reading

Do factor investing (smart beta) mutual funds capture for investors the premiums found in academic factor research? In their November 2022 paper entitled “Factor Investing Funds: Replicability of Academic Factors and After-Cost Performance”, Martijn Cremers, Yuekun Liu and Timothy Riley analyze the performance of funds seeking to capture of published (long-side) factor premiums. They group factor investing funds into four styles: dividend, volatility, momentum and q-factor (profitability and investment). They separately measure how closely fund holdings adhere to the long sides of academic factor specifications. They measure fund outperformance (alpha) relative to the market factor via the Capital Asset Pricing Model (CAPM) and via a multi-factor model (CPZ6) that accounts for the market factor and for granular size/value interactions. Using monthly returns for 233 hand-selected factor investing mutual funds and for the academic research factors during January 2006 (16 funds available) through September 2020 (207 funds available), they find that:

Keep Reading

Can investors make the stock return momentum effect stronger/more reliable by isolating stocks for which many similar stocks exhibit very strong or very weak past returns? In his December 2022 paper entitled “Neighbouring Assets”, Sina Seyfi explores this question by sorting stocks based on average past returns of other stocks with the most similar sets of 94 characteristics (neighbor stocks). He measures similarity between two stocks as the aggregate distance of their normalized and winsorized (excluding top and bottom 1% of values) characteristics over a baseline rolling 10-year history. His baseline “neighborhood” is 1,000 stocks. His baseline past return metric is average monthly value-weighted return of neighbor stocks over the past year. He considers three stock universes, consisting of all NYSE/AMEX/NASDAQ stocks: (1) excluding the 5% with the smallest market capitalizations; (2) excluding those below the 20% breakpoint of NYSE market capitalizations; and, (3) excluding those below the median of NYSE market capitalizations. He each month sorts stocks into tenths (deciles) of average past return of neighborhood stocks and reforms a value-weighted portfolio that is long (short) those in the decile with the highest (lowest) neighbor-stock average past return. Using monthly characteristics and returns for the specified stocks during January 1970 (with portfolio formation commencing January 1980) through December 2021, he finds that: Keep Reading

Do stocks with high exposures to new technologies outperform? In her December 2022 paper entitled “New Technologies and Stock Returns”, Jinyoung Kim examines future returns of stocks with relatively high exposures to new technologies as measured via patent analysis. Each June, she applies machine learning to both textual and citation information to detect technology areas with high growth in new patents and identifies new technologies based on the invention descriptions. For each U.S. public company, she then estimates the intensity of firm exposure to the last three years of new technologies. Excluding firms with zero exposure, she relates new technologies exposure to stock returns by each June reforming a portfolio that is long (short) stocks with the lowest (highest) 30% of new technologies exposures. Using information for all publicized U.S. patents and patent applications since 1976 (plus information for related pre-1976 granted U.S. patents and published international patents) and monthly returns for associated stocks and widely accepted stock factors during July 1981 through June 2019, she finds that: Keep Reading

Do the widely used U.S. stock return factors exhibit long-term trend changes and shorter-term cyclic behaviors? In his November 2022 paper entitled “Trends and Cycles of Style Factors in the 20th and 21st Centuries”, Andrew Ang applies various methods to compare trends and cycles for equity value, size, quality, momentum and low volatility factors, with focus on a breakpoint at the end of 2000. He measures size using market capitalization, value using book-to-market ratio, quality using operating profitability, momentum using return from 12 months ago to one month ago and low volatility using idiosyncratic volatility relative to the Fama-French 3-factor (market, size, book-to-market) model of stock returns. He each month for each factor sorts stocks into tenths, or deciles, and computes gross monthly factor return from a portfolio that is long (short) the average return of the two deciles with the highest (lowest) expected returns. As a benchmark, he uses the value-weighted market return in excess of the U.S. Treasury bill yield. Using market and factor return data from the Kenneth French data library during July 1963 through August 2022, he finds that:

Keep Reading

Is market capitalization, shares outstanding times share price, really the total value of a firm? In his brief November 2022 paper entitled “The Market Capitalization Illusion”, J.B. Heaton examines the relationship between market capitalization and market value considering the slope of the demand curve for tradable assets. Based on the body of relevant research, he concludes that: Keep Reading

In response to the U.S. stock market timing backtest in “Usefulness of P/E10 as Stock Market Return Predictor”, a subscriber suggested combining a 10-month simple moving average (SMA10) technical signal with a P/E10 (or Cyclically Adjusted Price-Earnings ratio, CAPE) fundamental signal. Specifically, we test:

• SMA10 – bullish/in stocks (bearish/in cash) when prior-month stock index level is above (below) its SMA10.
• SMA10 AND Binary 20-year – in stocks only when both SMA10 and P/E10 Binary 20-year signals are bullish, and otherwise in cash. The latter rule is bullish when last-month P/E10 is below its rolling 20-year monthly average.
• SMA10 OR Binary 20-year – in stocks when one or both of the two signals are bullish, and otherwise in cash.
• NEITHER SMA10 NOR Binary 20-year – in stocks only when neither signal is bullish, and otherwise in cash.

We use Robert Shiller’s S&P Composite Index to represent stocks. We consider buying and holding the S&P Composite Index and the standalone P/E10 Binary 20-year strategy as benchmarks. Using monthly data from Robert Shiller, including S&P Composite Index level, associated dividends, 10-year government bond yields and values of P/E10 as available during January 1871 through September 2022, we find that:

Keep Reading

In response to the U.S. stock market timing backtest in “Usefulness of P/E10 as Stock Market Return Predictor”, a subscriber suggested a modification for exploiting P/E10 (or Cyclically Adjusted Price-Earnings ratio, CAPE). Instead of binary signals that buy (sell) stocks when P/E10 crosses below (above) its historical average, employ a scaled allocation to stocks that considers how far P/E10 is from average. Specifically:

• If P/E10 is more than 2 standard deviations below its past average, allocate 100% to the S&P Composite Index.
• If P/E10 is more than 2 standard deviations above its past average, allocate 0% to the S&P Composite Index.
• If P/E10 is between these thresholds, allocate a percentage (ranging from 100% to 0%) to the S&P Composite Index, scaled linearly.

To investigate, we backtest this set of rules. Using monthly data from Robert Shiller, including S&P Composite Index level, associated dividends, 10-year government bond yields and values of P/E10 as available during January 1871 through September 2022, we find that:

Keep Reading

Does P/E10 (or Cyclically Adjusted Price-Earnings ratio, CAPE) usefully predict U.S. stock market returns? Per Robert Shiller’s data, P/E10 is inflation-adjusted S&P Composite Index level divided by average monthly inflation-adjusted 12-month trailing earnings of index companies over the last ten years. To investigate its usefulness, we consider in-sample regression/ranking tests and out-of-sample cumulative performance tests. Using monthly values of P/E10, S&P Composite Index levels (calculated as average of daily closes during the month), associated dividends (smoothed), 12-month trailing real earnings (smoothed) and interest rates as available during January 1871 through September 2022, we find that: Keep Reading