factor · Spillover

Spillover factor — explained

Spillover bets that news affecting a major customer or supplier predicts the linked ticker's next-month return. The market processes upstream/downstream relationships slowly — analysts cover one company in a chain, the other gets re-priced with a delay. The factor reads the lagged information that hasn't crossed the link yet.

Where this comes from

Academic anchor

Cohen-Frazzini 2008 — Economic Links and Predictable Returns
Documents that a portfolio long stocks whose principal customers gained the prior month and short stocks whose customers lost earned ~14% annually over 1980-2004 with no exposure to standard factors. The mechanism: investors track the company they cover but ignore the supplier 1-2 nodes upstream; news at the customer level only flows to the supplier when the supplier reports its own quarter. Menzly-Ozbas 2010 generalised the result to industry-level upstream-downstream networks (energy → chemicals → industrials). The lag is roughly 30 trading days — long enough to harvest, short enough to not need expensive optionality.
Plain English

What it actually measures

Apple reports a great quarter. Foxconn — Apple's primary assembly partner — does not move that day, even though 70% of Foxconn's revenue depends on Apple. Why? The analysts covering Apple write up Apple's report; the analysts covering Foxconn wait for Foxconn's own earnings call three weeks later. Between those three weeks, the new information about Apple demand sits in Foxconn's price as latent re-rating. Spillover formalises this — for every ticker we maintain a graph of revenue-significant suppliers and customers, and whenever a linked node moves, we propagate a shrunken version of the move to the dependent ticker.

No calibration constants

Math sketch

inputs   · customer-supplier link graph from 10-K disclosure
         · revenue-share weights per link
         · idiosyncratic peer return moves (peer return minus market)
         · time-since-news decay
ideas    · sum of weighted, time-decayed peer moves over the link graph
         · standardised across the universe so links are comparable
output   · cross-sectional standardised score

The link-graph extraction (which 10-K paragraphs we mine, how we score concentration), the time-decay function, and the post-news window length are calibrated and proprietary. Public: that we use a customer-supplier graph and that the academic anchor is Cohen-Frazzini 2008.

Pipeline

How Framler implements it

The link graph rebuilds quarterly from 10-K filings via a customer-segment XBRL tag where companies disclose it, plus a fallback heuristic that infers links from peer-comparison disclosures. Node moves are computed nightly from EOD prices. The factor lives in apex_factor_scores and refreshes during the 06:00 UTC universe sweep. For tickers with no disclosed links — common for diversified holding companies and conglomerates — spillover defaults to a neutral 50 (no link information available).

One coherent posterior

How it composes with Framler

Spillover composes with Sector momentum (Moskowitz-Grinblatt 1999) — when a sector is rotating up AND a key supplier is propagating positive flow, the SUPPLY-CHAIN MOMENTUM confluence pattern fires bullish. The opposite — sector rotating down plus negative spillover from a major customer — fires the SUPPLY-CHAIN CRACK pattern. Spillover also amplifies PEAD when the customer of a beat-stock is in our universe — the customer's earnings beat tomorrow is our spillover-positive bet today.

Honest limitations

When it fails

Two structural issues. (1) Graph completeness. We only see links companies disclose, which is a SEC-mandated subset (10% revenue concentration thresholds). Real supplier networks are dense and largely opaque. The mid-cap chemicals supplier with seven 8%-revenue customers is invisible to us — its actual diversification profile is masked because no single customer triggers disclosure. (2) Globalisation drift. The Cohen-Frazzini study was US-equity-only over 1980-2004. Modern supply chains span continents — a US ticker's 'real' customer concentration may be a Korean memory manufacturer that doesn't appear in any US-filed 10-K. Our coverage of cross-border links is patchy and on the post-16-May roadmap to extend.

Read next

Related factors

SectorPEADMomentum

See Spillover score on a real ticker

Every ticker page shows the per-factor decomposition. The Spillover score is one of thirteen composing the 0–100 the composite score.

Try NVDA →Full methodology59 invariants live
Spillover factor explained — Framler | Framler