Repository navigation
Trade Check and Sentiment
How a Buy or Sell becomes a verdict. Everything on this page is deterministic code running in AWS Lambda; no model decides a verdict or a number.
flowchart LR
A["Market change<br/>price · volume · breadth"] --> S
B["Headlines<br/>sentiment + scope<br/>pre-scored by Bedrock"] --> S
V["Ticker's own volatility<br/>60-day typical move"] --> G
P["Investor profile<br/>risk band · growth style"] --> G
S["Scores<br/>fundamental · emotion · hype"] --> D
G["Gates<br/>size · volume · unusualness"] --> D
D["Verdict + evidence IDs"]
| Signal | Meaning |
|---|---|
| Drop / rise | Trough or peak versus the last close before the refresh |
| Volume ratio | Peak volume versus the typical (median) volume |
| Typical daily move | The ticker's own 60-day volatility; the move is also expressed in these units |
| Breadth | Share of tracked tickers that also fell sharply |
| Company share | Share of the move that is company-specific: from headline scope when there are headlines, otherwise from the price data against the market-wide median move |
| Sentiment | Headline tone from -1 to +1, plus market-wide news tone |
| Score | Formula | Reads as |
|---|---|---|
| Fundamental | 0.5 × company share + 0.3 × negativity of sentiment + 0.2 × (1 − rebound) | The move is about the company and lasting |
| Emotion | 0.4 × (1 − company share) + 0.3 × breadth + 0.3 × volume intensity | The move is market- or crowd-driven |
| Hype | 0.4 × positive sentiment + 0.3 × volume intensity + 0.3 × size of rise | A rally fuelled by enthusiasm |
net = fundamental − emotion decides company-driven (≥ +0.15), market-driven (≤ −0.15) or mixed.
A move is flagged only if it passes all three checks. The bar rises with the investor's risk band, which already includes the account rule (an RRSP lowers it one step).
| Risk band | Minimum move | Volume spike | Unusual for the ticker |
|---|---|---|---|
| Low (e.g. Robert) | 2% | 1.3× normal | 1.5× its typical daily move |
| Mid (e.g. Mateo) | 3% | 1.5× normal | 2× |
| High (e.g. Amira) | 4% | 1.8× normal | 2.5× |
This is why a 2% drop on a volatile stock is not flagged, while the same 2% on a calm index fund can be.
| If you sell | Meaning |
|---|---|
panic_risk |
Sharp drop on heavy volume, market- or crowd-driven |
evidence_supported |
Sharp drop, the data backs the concern (company-driven, or consistent with your plan) |
mixed |
Signals disagree |
no_signal |
Too small, too ordinary, or volume too normal |
| If you buy | Meaning |
|---|---|
chasing_risk |
Sharp rise with upbeat headlines (buying the hype) |
falling_knife |
Sharp drop driven by negative company news |
dip_with_context |
Sharp drop driven by market-wide news |
mixed · no_signal
|
As above |
For a sell, the profile softens how a market-driven drop is read:
| Adjustment | Low | Mid | High |
|---|---|---|---|
Added to net (shorter horizon, less room to wait out a drop) |
+0.40 | +0.20 | 0 |
| Extra for a growth-style holding (growth or momentum, not dividend or preservation) | +0.30 | +0.10 | 0 |
Selling a growth stock does not fit a short-horizon, capital-preservation plan, so it is a reasoned decision, not fear. Buys keep the data's own reading.
A sell of a big holding is a big decision, so it must clear a higher bar before it is called reasoned. From 25% of the portfolio, the Classifier lowers net by 0.10, rising linearly to 0.25 at 50% (formula trade-guard-v4). It applies to sells only, and the holding's weight is shown as the evidence item position_weight.
Amira, the high-risk persona, deliberately holds only three stocks: NVDA about 42%, SHOP.TO about 28% and QQQ about 24%. Selling NVDA after a fear-driven drop lowers net by about 0.20, the pop-up says the position is a large share of her portfolio and suggests trimming in small steps, and the stance is careful. A sale backed by company-specific evidence (ENB.TO) still reads as reasoned. Buying more of a holding that is already 25% or more of the portfolio also adds a concentration note.
| Ticker | Move | × typical day | Robert (low) | Mateo (mid) | Amira (high) |
|---|---|---|---|---|---|
| NVDA | -11.4% | 4.2 | reasoned | mixed | panic |
| SHOP.TO | -9.5% | 2.8 | reasoned | panic | panic |
| QQQ | -6.8% | 4.5 | reasoned | panic | panic |
| SPY | -4.9% | 5.2 | mixed | panic | panic |
| XIU.TO | -2.3% | 2.6 | mixed | no signal | no signal |
| ENB.TO | -16.6% | 16.1 | reasoned | reasoned | reasoned |
Example scenario on simulated data, to show behaviour. It is not a forecast.
- The same inputs always give the same verdict.
- Every figure is an evidence item with an ID the Explainer may cite and nothing else.
- Verdicts describe how the data looks; they are not advice or a prediction.