PG-MIM™ Case Studies

The PG-MIM algorithm would have begun generating bearish signals as early as late January 2020, when the WHO declared COVID-19 a PHEIC. At that point, the Event Impact Score (EIS = -85) would have registered the event as a high-severity (0.9), global-scope (1.0) health crisis, producing a strongly negative EIS. The event category "geopolitical" carries a type weight of 1.0, amplifying the signal. With time-decay (7-day half-life), the most recent events (March circuit breakers) dominated the score while the January WHO declaration had decayed to approximately 15% weight.

The Exposure Score (ES = -78) flagged broad market vulnerability because the pandemic affected all sectors simultaneously, with particularly high exposure for travel, hospitality, energy, and retail. The geographic scope was global (G_scope = 1.0), meaning no sector or region was insulated. The sector sensitivity matrix rated the S&P 500's blended exposure to a geopolitical event at 0.6-0.95 across its constituent sectors.

The Sentiment Shift Score (SSS = -82) captured the rapid deterioration in market narrative. Media amplification was extreme, with 24/7 pandemic coverage driving the amplification factor well above normal levels. The shift from complacency (S_previous near 0) to panic (S_currentnear -1.0) produced a large negative SSS. Source credibility weighting was high (WHO, CDC, major financial outlets), amplifying the signal's contribution.

Critically, the Market Confirmation Score (MCS = +72) operated under the v2.0 divergence-based methodology. In late February, when the PG algorithm first generated its bearish thesis, the S&P 500 was still near all-time highs. The divergence between the strongly negative event thesis and the flat-to-positive price action produced a high positive MC score, indicating that the bearish thesis had NOT yet been priced in. This is the key innovation of the divergence-based MC: rather than waiting for price to confirm the thesis (which arrives too late to be actionable), the algorithm detects the opportunity window where the event signal exists but the market has not yet reacted. The +72 MC score amplified the overall bearish signal, boosting conviction precisely when it was most valuable, before the crash.

The Macro Compatibility Score (MaCS = -65) turned negative when the Fed executed an emergency 50bp rate cut on March 3, signaling that the central bank itself viewed economic conditions as deteriorating. Emergency rate cuts historically correlate with recessions, reinforcing the bearish macro environment. The macro event category weight of 0.9 ensured this signal contributed meaningfully.

The Uncertainty Score (US = 88) was elevated because the pandemic was an unprecedented event with no modern historical precedent, reducing model fit (F_model). Conflicting sentiments (some events positive, like the tariff delay) added +25 to uncertainty. However, the sheer magnitude and consistency of all five directional sub-scores overwhelmed the uncertainty discount. The raw PG Score of -68 was then processed through EMA smoothing (half-life = 3), which accumulated the persistent bearish signal over multiple periods to produce a smoothed score of -72. The recommendation hysteresis (margin = 5 points) ensured the Strong Sell signal remained stable once triggered, preventing false flips during brief relief rallies.

Multi-horizon validation showed the signal was most accurate at the 10-day horizon (78% directional accuracy), meaning the PG algorithm's bearish thesis was best validated by looking at price movement 10 trading days forward. The 1-day horizon showed lower accuracy (52%) due to intra-day noise, while the 20-day horizon showed 71% accuracy, confirming the signal's medium-term predictive power.

The PG-MIM algorithm would have begun accumulating bullish signals for NVDA as early as December 2022, following the launch of ChatGPT. The Event Impact Score (EIS = +82) classified the emergence of generative AI as a high-magnitude (0.85), global-scope corporate and technology event. While the initial EIS was moderate (the event was still speculative), it strengthened dramatically after NVDA's February 2023 earnings confirmed that AI demand was translating into actual revenue. With the 7-day half-life decay, the May earnings event dominated the score while earlier events contributed at reduced weights.

The Exposure Score (ES = +90) was exceptionally high for NVDA because the company held a near-monopoly position in AI training GPUs. The sector sensitivity for Technology to a corporate event was rated at 0.7, but NVDA's specific position as the dominant supplier pushed the direct-symbol impact multiplier to 1.5x, producing the highest exposure score of any case study.

The Sentiment Shift Score (SSS = +85) captured the narrative transformation. Before ChatGPT, NVDA was primarily viewed as a gaming GPU company with a growing data center business. After ChatGPT, the narrative shifted to "the picks-and-shovels play of the AI revolution." The shift from neutral sentiment (S_previous near 0) to strongly bullish (S_currentnear +0.9) produced a large positive SSS. Source credibility was high (SEC filings, major financial outlets), amplifying the signal.

The Market Confirmation Score (MCS = +58) under the v2.0 divergence-based methodology was moderately positive. In the weeks before the May 24 earnings report, NVDA had rallied but had not yet priced in the full magnitude of the AI demand thesis. The divergence between the strongly positive event thesis (driven by hyperscaler capex commitments and the ChatGPT adoption curve) and the still-moderate stock price produced a positive MC, indicating remaining upside opportunity. After the 24.4% post-earnings surge, the MC score would have declined as the price caught up to the thesis, but the EMA-smoothed score captured the pre-earnings opportunity window.

The Macro Compatibility Score (MaCS = +55) was moderately positive. While the Fed was still in a rate-hiking cycle (which typically pressures growth stocks), the AI spending boom represented a secular trend that transcended the monetary policy cycle. Hyperscaler capital expenditure announcements from Microsoft, Google, and Amazon provided macro-level confirmation.

The Uncertainty Score (US = 35) was relatively low because the earnings data provided concrete, quantifiable evidence. The 53% revenue guidance beat was not speculative; it was a hard number from the company's own financial reporting, giving the model high data quality and strong model fit against historical patterns of earnings-driven rallies. Low uncertainty translated to high confidence (0.82).

The raw PG Score of +74 was processed through EMA smoothing, which accumulated the persistent bullish signal from the ChatGPT launch through the earnings confirmation, producing a smoothed score of +78. The recommendation hysteresislocked in the Strong Buy signal, preventing false downgrades during minor pullbacks. Multi-horizon validation showed the signal was most accurate at the 20-day horizon (85% directional accuracy), reflecting the secular nature of the AI demand thesis, which played out over weeks and months rather than days.

The US-China trade war of August 2019 represents the ideal Hold scenario for the PG-MIM algorithm: a period where contradictory signals from multiple event categories cancel each other out, and the resulting high uncertainty appropriately suppresses confidence in any directional call. The v2.0 algorithm's EMA smoothing and hysteresis features were particularly valuable here, preventing the whipsaw news cycle from generating false signal flips.

The Event Impact Score (EIS = -15) was mildly negative because the tariff escalation (a geopolitical event with W_type = 1.0) was partially offset by the tariff delay announced on August 13. The 7-day half-life decay meant that the most recent events (the August 23 retaliation and August 26 resumed talks) had the highest weights, and since they pointed in opposite directions, they largely cancelled. The geographic scope was regional (US-China bilateral), not global, which reduced the G_scope multiplier to 0.7.

The Exposure Score (ES = -12) was moderately negative for trade-sensitive sectors (technology, industrials, agriculture) but near-zero for domestic-focused sectors (utilities, healthcare). The broad S&P 500 exposure was diluted because the tariff impact was concentrated in specific supply chains rather than affecting the entire economy uniformly.

The Sentiment Shift Score (SSS = +5) was the most telling indicator of the Hold signal. It oscillated between negative and positive multiple times within the month. On August 5, sentiment was deeply negative (yuan devaluation fears). By August 13, it had reversed to positive (tariff delay relief). By August 23, it was negative again (retaliation). By August 26, it was positive once more (resumed talks). The net SSS was near zero, reflecting the whipsaw nature of the news cycle. The EMA smoothing filtered out the rapid oscillations, producing a stable near-zero sentiment reading.

The Market Confirmation Score (MCS = -6) under the v2.0 divergence-based methodology was near zero. The S&P 500's 1.8% monthly decline roughly matched the mild bearish thesis (EIS = -15), meaning there was no significant divergence between the event thesis and price action. A near-zero MC indicates the thesis is approximately priced in, neither an opportunity window (positive MC) nor a caution signal (strongly negative MC). This is the correct behavior for a Hold scenario: the market has already absorbed the available information.

The Macro Compatibility Score (MaCS = -10) was mixed. The yield curve inversion on August 14 (a historically reliable recession predictor) pushed MaCS negative. However, employment data remained strong, consumer spending was healthy, and GDP growth was positive, which partially offset the inversion signal.

Crucially, the Uncertainty Score (US = 82) was very high because the contradictory signals from multiple event categories reduced model fit significantly. The presence of both positive and negative sentiment events triggered the +25 conflicting-signal penalty. When the algorithm cannot establish a consistent directional thesis across its sub-components, it correctly raises uncertainty and lowers confidence. The resulting raw score of -5 was smoothed to -3 by the EMA, and the recommendation hysteresisensured the Hold signal remained stable throughout the month despite the intra-month volatility. The low confidence of 0.38 meant the PG Signal™ engine correctly issued a Hold rather than attempting a directional call.