Chapter 3: A tour of risk and performance

• Alpha is the expected value of the idiosyncratic return and epsilon is the noise masking it.

  • There's general agreement that accurately forecasting market returns is very difficult or at least not the mandate of a fundamental analyst. A macro investor may have an edge in forecasting the market, but the fundamental one does not have a differentiated view.
  • The job of the fundamental investor is to estimate alpha accurately
  • The risk manager is job is to estimate beta and identify the correct benchmark

• The error around an alpha estimate is larger than that alpha estimate itself. This is not true for beta. You cannot observe alpha directly and you cannot estimate it from a time series of returns. The fundamental analyst predicts forward looking alphas based on deep research. The ability to combine these alpha forecasts from a variety of sources and process a large number of unstructured data is a competitive advantage of the fundamental investor

• The chapter decomposes alpha(idio) and beta contribution and risks

  <aside> 📌 [Kris: found this helpful for computing portfolio alpha I’ve covered this all before in https://moontowermeta.com/from-capm-to-hedging/

  • What the industry calls “idio”, I refer to as “risk remaining”. Same exact concept.

  • If X = market and Y = stock then this identity decomposes the risk:

    • Stock variance = Market variance + Idio variance

      <aside> ➡️ Var (Y) = $R^2$ * (VarY/VarX) + (1-$R^2$) * (VarY/VarX)

      </aside>

      Note that market variance is $B^2$

      Recall that B = R * ratio of volatility between Y and X

  • Risk remaining or idio is a non-linear function of correlation. As correlation drops, idio explodes. Once R is .86 or lower the idio risk exceeds the market risk!! </aside>

• Shorting SPY to hedge out beta and have absolute return dependent strictly on idio

Chapter 4: An introduction to multi-factor models

• Bar Rosenberg was an academic who started Barra, now MSCI, BGI (acquired by Blackrock), and Axa Rosenberg.
• Since 1981 there has been an explosion of factors explaining equity returns. The methodologies were later extended to other assets like government and corporate credit.
  • How are the factors discovered? And where do the betas come from? There are three broad ways to attack the problem.
    1. The time-series approach, most popular in academia, is often used in conjunction with the fundamental approach. We are surrounded by time-series data: macroeconomic indicators, inflation, interest rates, economic growth, employment data, and business cycle indicators. Economic sentiment data, like consumer confidence, and detailed economic activity data such as same-store sales, purchasing manager indices, and international vs. domestic activity, are also included. Commodity prices, especially oil, also affect economic activity. All of these data are factor candidates. The sensitivity (or beta) of a stock to these factors is estimated by performing a time-series regression of the stock return to the time series. The factor returns (the time series) are known; the betas are derived using asset and factor returns.
    2. The fundamental approach (the most popular with fundamental managers — easy to interpret). This is the method pioneered by Barr Rosenberg. Each stock has characteristics. These characteristics describe the company and its returns. We have to identify the relative Characteristics, which serve as the betas of the model. We. then need to populate these fields for every stock, every characteristic, and every period. In this model, the betas are the primitive information, the factor returns are derived: given the betas and the asset returns, you can estimate the factor returns.
    3. The statistical approach: In this method, you are given only the time series of asset returns. Both the beta and the factor returns are estimated using the asset returns alone. This may seem magical. Don't we need one or the other? The approach sidesteps the problem by choosing betas and returns so as to describe most of the variation in the asset returns. This is the hardest model to interpret and is often used as a "second opinion" to check whether the base model is missing some systematic source of risk.
• A wave metaphor to understand factor models
  • A large wave is composed of fewer, smaller waves riding on it, with many ripples on top of the smaller waves. The effects accumulate. Similarly, stock returns are the result of a large shock, such as the market, followed by a few smaller ones, like sectors or larger style factors, and then even fewer, smaller ones. Like waves, many of these movements have similar amplitude over time. We say that their volatility is persistent. This is a blessing because it allows us to estimate future volatilities from past ones. However, in some cases, the volatility of a factor is not very persistent. Examples include short interest, hedge fund holdings, or momentum. You can think of these large factor returns as a tsunami that will wreak havoc on your portfolio if you are not prepared for it.

Chapter 5: Understanding Factors.

• The interpretation of an industry factor's return differs from the return of a simple, cap-weighted portfolio of stocks in that industry. As in the case of sector SPDRs, the interpretation of the risk model's industry portfolio is more involved. It has unit exposure to that industry and no exposure to any other industries, like an industry benchmark. In addition, it has no exposure to any style factor, whereas the Technology SPDR ETF may have, at some point in time, a positive momentum exposure because all of its members have enjoyed positive returns. An industry media factor portfolio will have no momentum exposure. The benefit of the factor-based approach is that its returns are not conflating the returns of several factors. You know that internet media has positive returns not because of its momentum content.

• The term structure of the momentum factor:

  • Strong (weak) performance in the previous month is reverting. The effect is stronger as the historical observations are shortened. Strong performance in the past week reverts more than strong performance in the past month.
  • Performance in the interval between one month and one year shows continuation, i.e., true momentum.
  • Performance beyond a year is reverting. A stock with large positive returns in the period between 15 months ago and a year ago will have negative expected returns, all other characteristics being equal.

  Interpretation

  • We don't understand the origins of momentum. We have complex, hard-to-falsify theories about overreaction and underreaction.
  • A risk-based explanation suggests that momentum returns exhibit a heavier left tail. In fact, some studies employ a left tail dependence factor and argue that momentum is a redundant factor.

Chapter 6: Use effective heuristics for alpha sizing

• RenTech had identified alpha signals, but it took years until they would produce sizable revenues. It was only in the new millennium that its sharp ratio went from an already good 2 to an outstanding 6 and higher. What made this qualitative jump possible was portfolio construction, a process that takes many inputs and generates a single output (the portfolio itself). Portfolio construction is a competitive advantage for quantitative managers, if not the competitive advantage. (alpha sizing as we will see can produce significant differences in SRs)
• The daily job of a fundamental portfolio manager is a continuous learning process. The better the process, the faster the learning.
• A few heuristics without any optimization will be sufficient for 80 to 90% of portfolio managers. (at least until returns shrink because of market efficiency or higher costs. or the cognitive cost of managing a portfolio. increases dramatically, usually as your coverage increases).
• Required inputs:
  • Expected returns (Kris: edge)
  • Risk estimate
  • guidelines on strategy’s maximum risk
• In practice sharp ratio usually does not subtract the risk free rate. Despite its shortcomings, it is still useful.