1.10 Factor Investing & Factor Timing

Basic Prerequisites: Econometrics & Time Series (regression, covariance, stationarity) and Fundamental Multi-Factor Models (how the factors are built - this folder assumes that and asks what happens when you trade them). (these are the folder-level prerequisites for pages 02–06; page 01 states its own, smaller, entry requirements)

1. Intuition & Practical Objective

A factor is a small, persistent, compensated return that is orthogonal to the market: value (cheap beats expensive), momentum (winners beat losers), quality (profitable/low-accrual beats junk), low-volatility (low-beta beats high-beta on a risk-adjusted basis), and size (small beats large). Each is an anomaly in the sense that it is not explained by the CAPM’s single beta - Fama and French (1992) showed the cross-section of average returns lines up with size and book-to-market, not with market beta. Cochrane (2011) calls the resulting proliferation of such return-predictive signals, in the title of his presidential address, a “zoo of new factors” - and asks the practitioner’s question directly: how many of them are really important?

This folder is the investing and practice half of that story. Its sibling, Fundamental Multi-Factor Models, covers the machine: how SMB/HML/RMW/CMA are constructed by 2×3 sorts, how you estimate loadings by time-series regression, and how a Barra/Axioma cross-sectional model fits factor returns each period. This folder asks what happens next: once you have a factor, how do you invest in it, how much capacity does it have, what does publication do to its premium, can you time it, and what kills it.

It is a hub: it (a) gives the practitioner lookup below (premiums, IC, the timing signal, the capacity formula), and (b) routes you to six sub-pages that walk from raw intuition through the factor zoo, crowding and capacity, post-publication decay, the failure modes of practice, and the extensions - factor timing.

The one-sentence essence. “A factor is a traded version of a cross-sectional anomaly whose premium is an arbitrage with a half-life: it decays as capital crowds in (investors learn, publish, and compete), and the practitioner’s whole job is to size it for capacity and to decide whether any of that decay is predictable (factor timing) - while remembering that timing is a high-variance, low-IC bet.”

Audience arc: the beginner learns why one beta fails and what a factor premium actually is; the intermediate learns the factor zoo, its multiple-testing problem, and how publication and crowding erode a premium; the expert reads capacity estimation via square-root market impact, McLean–Pontiff post-publication decay estimation, and the Grinold fundamental law applied to factor timing. The sub-pages below are ordered for exactly that climb.

2. Mathematical Ground Truth & Practitioner Lookup

Quick lookup. The folder’s key formulas in one table; each is derived on the sub-pages.

Notation: $R_{it}$ asset $i$ return at $t$; $f_t$ a factor’s long-short return; $C_{it}$ security $i$‘s characteristic (standardized); $N$ cross-section size; $K$ number of candidate factors; $I{C}_t$ the cross-sectional information coefficient; $S_t$ a factor-timing signal; $A$ assets under management; $Q$ traded quantity; $ADV$ average daily volume.

The cross-section of expected returns. The panel-forecasting view (Cochrane 2011, §II):

A portfolio sort is a nonparametric cross-sectional regression with non-overlapping histogram weights (Cochrane Fig. 7): sorting on one characteristic and reading the decile-1-to-decile-10 mean spread is the slope estimate, just in a different functional form. The 1–10 information ratio equals the Sharpe ratio of the underlying factor, equals the $t$-statistic of the cross-sectional regression coefficient.

The information coefficient and the fundamental law (Grinold 1989; Grinold & Kahn). The IC is the cross-sectional correlation between the characteristic and subsequent returns, $I{C}_t={\mathrm{corr}}_i(C_{it},R_{i,t+1})$, and the expected information ratio of a score-weighted portfolio is

This single equation is why a weak cross-sectional signal (IC ≈ 0.03–0.05) still makes a good long-short factor (breadth ≈ 12 × 300 stocks), and why factor timing is hard: timing has breadth of order 12 (one bet per month on the factor), so an IC of 0.05 buys an IR of only $0.05\sqrt{12}\approx 0.17$.

The factor-timing signal. The two canonical timing signals are (i) the valuation spread - how expensive the factor is versus its own history, $z_t=(S_t-\bar{S})/{\sigma }_S$, which predicts higher factor returns when the spread is wide (the factor is “cheap”); and (ii) trend on the factor itself, $z_t=\frac{1}{L}{\sum }_{j=1}^L{f}_{t-j}$, which predicts continuation. The timed factor return is

and its appraisal ratio is again $IC\times \sqrt{\text{breadth}}$ - the same law, with much smaller breadth.

Capacity under square-root market impact. With gross alpha $g$ per year, turnover $\tau$ (round-trips/yr), and one-way impact modelled as

the net alpha at scale $A$ is

Net alpha is concave-decreasing in AUM and hits zero at the capacity break-even - the empirical estimate in §3 is $\approx$$630bn for a large, liquid, 4%/yr factor turning over twice a year.

Post-publication decay (McLean & Pontiff 2016). Normalize each characteristic’s out-of-sample and post-publication mean return by its in-sample mean:

MP find ${\hat{d}}_{\text{OOS}}\approx 10\%$ (not statistically different from 0) and ${\hat{d}}_{\text{post}}\approx 35\%$ (highly significant, $t\approx -4.9$ in their pooled specification). A 5% in-sample alpha “is expected to decay to 3.25% post-publication.”

The multiple-testing hurdle (Harvey, Liu & Zhu 2016). If $K$ factors are tested independently at the 5% two-sided level, the expected number of false discoveries is $0.0455K$; the Bonferroni threshold for $K=300$ is $|t|>3.76$, and HLZ argue for a $t>3$ hurdle in the factor literature. This is the statistical face of the factor zoo.

3. Computational Implementation - the practitioner engine

This runs on the standard library only (no numpy/scipy). It does three things every factor practitioner does: (1) sort a cross-section, form the long-short spread, and read off the information coefficient; (2) run the multiple-testing experiment that produces the zoo; and (3) estimate the capacity break-even.

The spread recovers the true 0.30%/month per-unit premium - a decile spread spans ≈3.5 s.d. of a normal characteristic, and $0.003\times 3.51=1.05\%$/mo, which is what the simulation delivers. The IC is a modest 0.045, and the long-short Sharpe is 2.05 - the entire case for factor investing in three numbers: a weak signal, applied across a wide cross-section, produces a strong diversified portfolio. Breadth is doing the work, not signal strength.

(Every number on every sub-page was re-executed and reproduced exactly from this engine and its siblings; the full build-out is on the sub-pages.)

4. Failure Modes & First-Principles Breakdowns

Hub signposts - the folder’s failure-mode analysis lives in 05 · Failure Modes & Practice. In one line each:

1. Factor crowding & liquidity black holes - when many funds hold the same long-short, a forced unwind has no counterparty; realized tails are far wider than the modeled beta-risk (the Quant Quake of August 2007). Modeled $\Sigma$ assumes independent flow shocks and systematically understates this.
2. Post-publication decay - a premium is an arbitrage with a half-life; McLean–Pontiff measure a ~35% decay after publication, and the decay is largest for the cheapest-to-arbitrage characteristics (exactly the mispricing signature).
3. The factor zoo / multiple testing - with hundreds of candidates, ~4.6% look significant at $t>2$ by chance; the correct hurdle is $t>3$ (HLZ) or a Bonferroni/deflated-Sharpe correction (see Backtesting Hygiene & Deflated Sharpe).
4. Factor timing is a high-variance, low-IC bet - breadth is ~12, so timing needs an IC ≈ 0.1 to matter, and in-sample-optimal timing weights routinely collapse out-of-sample.

5. References

• Fama, Eugene & French, Kenneth: “The Cross-Section of Expected Stock Returns” (Journal of Finance, 1992)
• Cochrane, John H.: “Presidential Address: Discount Rates” (Journal of Finance, 2011)
• McLean, R. David & Pontiff, Jeffrey: “Does Academic Research Destroy Stock Return Predictability?” (Journal of Finance, 2016)
• Harvey, Campbell; Liu, Yan & Zhu, Heqing: ”… and the Cross-Section of Expected Returns” (RFS, 2016)
• Ilmanen, Antti: Expected Returns (Wiley, 2011)
• Grinold, Richard: “The Fundamental Law of Active Management” (JPM, 1989)
• Tsay, Ruey S.: Analysis of Financial Time Series (3rd ed.)

6. Connected Graph Bridges

• Foundational base: Econometrics & Time Series · Linear Algebra & Matrices
• Sibling topic (the construction half): Fundamental Multi-Factor Models - FF 3/5-factor construction, cross-sectional models, statistical factors
• The factor family: Cross-Sectional & Time-Series Momentum · Quantitative Fundamental Investing (the characteristics)
• Discipline: Backtesting Hygiene & Deflated Sharpe (multiple testing, DSR) · Regime Detection (regime-conditional factor behaviour)
• Downstream: Portfolio Optimization (factor-risk budgeting, capacity-constrained sizing) · Quantitative Risk (the factor covariance $\Sigma =B\Omega B^{\prime }+D$ - and its crowding blind spot)
• Sub-pages (in-folder): 01 From Zero · 02 The Factor Zoo · 03 Crowding & Capacity · 04 Post-Publication Decay · 05 Failure Modes & Practice · 06 Advanced Extensions (Factor Timing)

Beginner: start at 01 · Practitioner: start at 05