High-Frequency Data Triangulation
Combine independent live signals to estimate conditions before official data arrives
- Difficulty
- Advanced
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 95%
Define the outcome that delayed official statistics cannot yet measure, such as layoffs or consumer demand. Select multiple high-frequency indicators representing independent behaviors—for example, card usage, relevant search activity, cancellations, mobility, or transaction volume. Establish normal baselines, monitor changes in direction and intensity, and compare the signals rather than trusting any one proxy. A conclusion strengthens when several indicators with different biases converge on the same result. Analysts should then compare the provisional estimate with later official releases, measure where each proxy succeeded or failed, and update its weighting. The framework improves timeliness without pretending that alternative data is complete, representative, or immune to changing relationships during a crisis.
Origin
Extracted from Masters in Business, where Joe Davis explained Vanguard's use of credit-card activity, keyword searches, and cancellation behavior to assess pandemic conditions ahead of lagging reports.
Core principles
- 01Official economic statistics often arrive with a lag.
- 02No single alternative signal should determine the conclusion.
- 03Independent signals become useful when they converge.
- 04Search, spending, and behavioral data can reveal changes in near real time.
- 05Alternative indicators must be checked against the broader outlook.
How to run it
- 1
Define the target condition
Specify the economic or operational outcome that must be estimated before official reporting catches up. Define its geography, population, and time window.
Pro tip Choose a target narrow enough that candidate indicators have a defensible relationship to it.
Watch out A vague target makes unrelated signals appear falsely consistent.
- 2
Choose independent signals
Select several high-frequency data sources linked to different behaviors associated with the target. Document why each signal should move when the target changes.
Pro tip Mix transactional, behavioral, and search-based indicators when available.
Watch out Multiple feeds derived from the same underlying source are not independent confirmation.
- 3
Build baselines
Measure each indicator under normal conditions and account for seasonality, calendar effects, and structural differences. Express current movement relative to that baseline.
Pro tip Retain comparable historical disruptions when they exist.
Watch out Raw volumes can mislead when platform usage or coverage has changed.
- 4
Triangulate direction and timing
Compare whether the signals agree on the direction, onset, and severity of change. Investigate divergence rather than averaging it away.
Pro tip Give more weight to indicators with direct behavioral links and stable historical performance.
Watch out Correlation during one episode does not establish a durable causal relationship.
- 5
Calibrate against official data
When conventional reports arrive, compare them with the provisional estimate. Update indicator weights and document blind spots for the next cycle.
Pro tip Track revisions as well as first-release official figures.
Watch out Official data can also be incomplete or revised, so do not treat its first release as perfect ground truth.
In the wild
An analyst cannot rely solely on unemployment claims because phone lines and websites are overloaded. The analyst combines searches for unemployment terms, card-spending contractions, business closures, and mobility changes. Their shared direction indicates labor-market deterioration before complete claims are processed.
→ Leaders receive an earlier, qualified estimate of labor stress and can prepare support capacity sooner.
A retailer tracks card transactions, store visits, delivery searches, and product availability after a storm. When transaction and mobility signals recover but availability remains weak, the team identifies supply constraints rather than continuing demand weakness.
→ Inventory and logistics resources are directed to the actual bottleneck.
Common mistakes
Trusting one proxy
A single indicator may reflect platform-specific behavior, media attention, or measurement failure rather than the target condition.
Ignoring changing relationships
A proxy that worked in normal periods may behave differently when policy, access, or public behavior changes during a crisis.
Skipping retrospective calibration
Without comparing early estimates with later reports, analysts cannot improve indicator selection or weighting.
Is it for you?
Best for
It is best for analysts and decision-makers working during fast-moving disruptions where waiting for conventional reports is costly.
Not ideal for
It is not ideal when alternative signals lack stable relationships to the outcome or contain severe selection bias.
From the transcript
“But we're trying to triangulate.”
“When you look at I don't know the usage of of of credit cards, when you look at Google searches of of certain keywords, we…”
“I think you know, when you when you look at the distillation of of information and as well some more high frequency UH data, I…”
From the episode
Barry Ritholtz Discusses WFH’s Impact on the Economy (Podcast)