Event-driven market research
News moves markets.ShockSense measures how.
Turn financial headlines into structured events and compare them with how similar events were historically followed by the market.
Operating on a pilot historical dataset of 3128 validated events. ShockSense abstains when evidence is insufficient.
Detected
the anchor
Session close
exit
Operating on a pilot historical dataset of 3128 validated events. ShockSense abstains when evidence is insufficient.
Free account
Alerts for the sectors you follow.
Tell ShockSense which sectors and kinds of news you care about. It sends you the matching news as it happens, and the results after the close.
Research only, not investment advice.
- 01Pick what you followSectors and kinds of news: earnings, deals, regulation, company news, macro. Everything, or only clear signals.
- 02Instant, or once a dayAlerts as ShockSense detects the news, and a daily digest after the close with the results. On Telegram, by email, or both.
- 03Free, no passwordSign in with a code sent to your email. See everything we sent you, change your choices, or delete it all, any time.
Live
Latest events.
News that passed every check, as soon as ShockSense stored it: which way similar news leaned at that moment, and what the market did once the next close came in.
2,340 live events so far
The lean, scored780 of 1,535 right (51%)112 waiting for their close, next result after Nov 17, 16:00 ETScorecard
How ShockSense works
From headline to historical reaction.
One real event, followed through all three stages.
01 / 03
Classify
Headline → structured market event
Obvious noise is screened out before any AI call. Our classification model then extracts the event category, the company it is about and its ticker; the ticker must match the named company, or ShockSense abstains.
- Classify (current)
- Measure
- Compare
Headline
“Macerich Acquires Oakridge Mall in San Jose, CA for Approximately $250 Million”
- Event type
- M&A
- Ticker
- MAC
- Classifier confidence
- 85%
- Detected
- Nov 16, 08:07 ET
Detected
Nov 16, 08:07 ET
First observation
Nov 16, 09:30 ET · +82 min
Closing observation
Nov 16, 16:00 ET
Before the open
MAC raw return
SPY over the same window
Market-adjusted return
All eligible M&A events in the dataset
- Sample
- n=213
- Median
- −0.45%
- Consistency
- 60%
Historical data
Every observation, not a summary of them.
With a pilot-sized sample, averages hide more than they show. Each dot is one eligible event and its post-detection return.
Company-specific
No clear signal- Sample
- n=1,517
- Median
- −0.27%
- Consistency
- 56%
Company-specific
No clear signal- Sample size
- 1517
- Median
- −0.27%
- Directional consistency
- 56%
No clear signalEstablished sampleEarnings
No clear signal- Sample size
- 957
- Median
- −0.01%
- Directional consistency
- 51%
No clear signalEstablished sampleRegulation
No clear signal- Sample size
- 283
- Median
- +0.04%
- Directional consistency
- 51%
No clear signalEstablished sampleM&A
No clear signal- Sample size
- 213
- Median
- −0.45%
- Directional consistency
- 60%
No clear signalEstablished sampleInflation / Macro
No clear signal- Sample size
- 43
- Median
- +0.25%
- Directional consistency
- 63%
No clear signalEstablished sampleGeopolitics
No clear signal- Sample size
- 43
- Median
- −0.03%
- Directional consistency
- 53%
No clear signalEstablished sampleCommodity Shock
No clear signal- Sample size
- 29
- Median
- +0.03%
- Directional consistency
- 52%
No clear signalEstablished sampleMonetary Policy
No clear signal- Sample size
- 23
- Median
- −0.10%
- Directional consistency
- 65%
No clear signalEstablished sampleFiscal Policy
No clear signal- Sample size
- 20
- Median
- −0.03%
- Directional consistency
- 55%
No clear signalEstablished sample
Model portfolio
A simulated portfolio that only acts on evidence.
The ShockSense Model Portfolio follows qualifying signals with fixed rules, forward in time, and reports every result against SPY. When the evidence is not there, it waits.
Current simulated value
SIMULATED$102,962.86
- Model portfolio
- SPY total return
- Signals
- 318
- Active
- 0
- Closed
- 51
$100,000 simulated starting capital. Positions are generated from ShockSense's historical-bias framework. No real money is invested.
Methodology
The clock starts when ShockSense sees the news.
Post-Detection Return is measured from the first eligible market observation after detection through the close of the first full trading session after detection. Movement before detection is never counted as a reaction.
Published
Provider timestamp. Stored for provenance, never used as the anchor.
Detected at
When ShockSense received the headline. The only anchor.
First eligible observation
Open of the first hourly bar at or after detection — or the next official open.
Closing observation
Official close of the first full session after detection.
Raw
What the asset did.
The asset's return over the window, on actual traded prices, including dividends that go ex in the window.
MAC · M&A−1.93%
Market-adjusted
What the market didn't explain.
The asset's return minus SPY's return over the identical start and end observations.
adjusted = asset − SPY
MAC vs SPY−0.64%
Research integrity
Abstention is a feature.
Most headlines are not clean events, and most samples are too small to say anything. ShockSense keeps every observation traceable, excludes what can't be defended, and says so when the evidence is thin.
Evidence when it exists.
Abstention when it doesn't.
From raw observations to the validated pilot set
4,105raw3,128validated
Raw observations in the pilot database
4105
Hand-written sample events, not live detections
−12
4093
Recaps, transcripts, market wraps and price-move reports
−187
3906
Failed the event-quality rules
−55
3851
Ticker did not belong to the company in the headline
−3
3848
Excluded in the manual pilot review (stale, ambiguous or not an event)
−11
3837
No valid Post-Detection measurement
−204
3633
Duplicate of an earlier event
−505
3128
Validated pilot events
3128
Out-of-sample check
Raw returns
6
historical biases issued across 3128 walk-forward evaluations
Market-adjusted
7
historical biases issued across 3034 walk-forward evaluations
Each event is judged only against events whose measurement windows had already closed. With the pilot sample, no category clears the evidence thresholds — so ShockSense issues no bias at all.