Sentiment analysis (lens)

Definition

The sentiment lens evaluates what investors, analysts, media, and social participants believe and how they are positioned — as distinct from what the business or the price series itself says. Its premise is that beliefs and positioning can be measured, that they swing beyond what fundamentals justify, and that extremes carry information (usually contrarian, sometimes confirming).

How it works / structure

• Inputs: survey and positioning data (COT reports, short interest, fund flows), analyst activity (ratings, estimate revisions), options positioning (opt-put-call-ratio), and text-derived measures from news and social feeds.
• Core operations: normalize a raw measure against its own history (percentiles, z-scores), separate level from change, and classify readings as extreme or neutral; text pipelines score polarity and volume of coverage.
• Output shape: standardized readings (“short interest at the 90th percentile of its 3-year range”) that can be conditions in a falsifiable thesis.
• Concept-catalog mapping: positioning-type sentiment evidence binds to the flow family of the analysis-concept catalog.

When it applies

Most informative at extremes and around crowded trades: heavily shorted names, one-sided positioning in futures, unanimous analyst views. Useful on horizons of days to months. It is a weak standalone lens in calm, mixed-reading conditions, and it says nothing about what an instrument is worth — only about who already holds what view.

Risk profile & failure modes

• Extremes can extend: crowded is not the same as wrong; shorts can stay pressed and euphoria can persist longer than a thesis window.
• Measurement lag and revision: short interest is published on a delayed cycle and COT weekly, so readings describe positioning as of the report date, not today.
• Proxy validity: text-sentiment scores measure the text pipeline as much as the crowd; different vendors disagree on the same day.
• Reflexivity: once a sentiment reading is widely used as a contrarian signal, behavior around it changes.

Evidence & limits

Baker and Wurgler (2006) constructed a composite sentiment index and found that high sentiment predicted lower subsequent returns for hard-to-value, hard-to-arbitrage stocks — the canonical evidence that sentiment matters cross-sectionally. Tetlock (2007) found high media pessimism predicted short-horizon price pressure followed by reversion. Evidence for simple retail-facing gauges (put/call ratios, surveys) as standalone timing tools is mixed; treat any specific threshold rule (“ratio above X means Y”) as unproven unless a current study is cited. Popular sentiment folklore (magazine-cover indicator and similar) is folklore.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Short interest in X, above the 90th percentile of its 3-year range today, will fall below the 50th percentile within 90 days” — falsified if the published readings never drop below that mark.
• “Consensus FY EPS estimates for Y will be revised down at least 5% within two quarters” — falsified if consensus never falls 5% from today’s level in that window.

Cross-references

• Pillar 12 data entries: sent-short-interest, sent-cot-reports, sent-fund-flows, sent-analyst-revisions, sent-news-social
• Options positioning: opt-put-call-ratio, indicator-options-flow
• Behavioral foundations: bias-herding, bias-recency
• Adjacent lenses: lens-options (volatility as priced fear), lens-market (breadth as revealed participation)