How DeepCap Works

DeepCap uses machine learning to surface stocks with a higher‑than‑average probability of beating their regional benchmark with a positive return over the next 6 or 12 months. We combine quality fundamentals, momentum and trend, valuation, and technical confirmation to produce three actionable DeepLists.

How DeepCap Picks Stocks

Clear inputs and Calculated Probabilities — we transform high-quality market and company data into calibrated probability scores, then curate them into the three DeepLists you see on the subscribers' Dashboard.

Our process at a glance

Quality data acquisition

Daily prices & volume with corporate actions; fundamentals (ROE, EPS growth, margins, leverage); valuation (P/E, PEG, earnings yield); market context (index, sector, beta, volatility).

Cleaning & filtering

Split/dividend-adjusted price history; liquidity & price floors; currency normalization; survivorship-bias controls; completeness checks so weak data never reaches the model.

Feature engineering

Signals distilled into pillars: momentum & trend, quality & resilience, valuation, risk, and technical confirmation (e.g., 20/50/200-DMA, RSI/MACD regimes).

Machine learning & proprietary blend

Horizon-specific models (6M & 12M) estimate the probability a stock will beat its regional benchmark with a positive return. For each horizon, several gradient-boosted tree models are trained on successively longer stretches of history, each calibrated with isotonic regression on the period that follows, and their calibrated outputs are averaged into that probability.

From probabilities to curated lists

We apply guardrails (liquidity, diversification, data quality) and rank by the DeepScore described below.

What We Look At

Momentum & Trend
  • What we compute: 6/12-month momentum, relative strength vs. index, trend persistence and health.
  • Why it matters: Leaders often continue leading when trends are structurally sound.
Quality & Resilience
  • What we compute: ROE, EPS growth stability, margins, cash coverage (current ratio), prudent leverage.
  • Why it matters: Durable businesses better sustain advances across cycles.
Valuation
  • What we compute: P/E, PEG, earnings yield vs. sector norms.
  • Why it matters: Avoids names priced for perfection that can't meet embedded expectations.
Risk & Market Context
  • What we compute: beta, realized volatility, drawdown behavior, index/sector effects.
  • Why it matters: Provides the backdrop for persistence and portfolio fit.
Technical Confirmation
  • What we confirm: price vs. 20/50/200-DMA alignment, RSI zones, MACD regime.
  • Why it matters: Confirms that momentum is supported by market structure—not noise.

Vetting by Analysts

Before publication, DeepCap analysts review top candidates for liquidity traps, event risk (earnings gaps, guidance, M&A), governance/accounting flags, and other factors that may have slipped through. This step removes edge-case false positives while keeping the engine systematic and reliable.

From calibrated scores to the three DeepLists

Two forecast horizons, one overlap list, and continuous monitoring — expressed as a simple 0–100 DeepScore.

Model stack (simplified)

  • Separate objectives: Dedicated 6-month and 12-month models capture different market dynamics and class balance.
  • Gradient-boosted models: For each horizon, several gradient-boosted tree models are trained on successively longer stretches of history, on the same feature set.
  • Time-ordered evaluation: The method is evaluated by training on earlier periods and scoring a later period the training did not see.
  • Probability calibration: Each is calibrated with isotonic regression on the period that follows its training data, and the calibrated outputs are averaged into an estimated probability that the stock beats its regional benchmark with a positive return over the horizon.

Quality controls & guardrails

  • Liquidity & price floors to avoid illiquid, easily manipulated names.
  • Data integrity (completeness, corporate actions, fundamentals sanity checks).
  • Diversification guardrails to avoid over-concentration by sector/industry/region.
6M Momentum — tactical leadership
  • Optimizes for: recent momentum and trend health, confirmed technically.
  • Profile: faster rotation; useful as a tactical sleeve.
  • Refresh cadence: monitored continuously; typically refreshed monthly/quarterly.
12M Trend — steadier compounders
  • Optimizes for: financially stronger names with durable trends and fair valuations.
  • Profile: more selective; candidates often carry higher calibrated probabilities.
  • Refresh cadence: monitored continuously; typically refreshed quarterly/semi-annually.
High-Conviction — overlap of leaders
  • How it's formed: intersection of top-ranked names from both 6M and 12M after guardrails.
  • Intent: concentrate on names strong across timeframes.
  • Use case: focused satellite or upgrade source for a core allocation.

What the DeepScore means

DeepScore (0 to 100) is the score each DeepList is ranked by, with one score for each horizon. It starts from the model's calibrated probability for that horizon on a 0 to 100 scale. Two kinds of flag then adjust it: a distress or balance-sheet flag (for example negative book equity or three years of operating losses) sets it to zero, and buybacks alongside falling earnings reduce it by a tenth. Stocks that fail the investability filters cannot enter a DeepList, and their published score is held at or below the lower quartile of scores.

Analyst review, publication & monitoring

  • Human review: news flow, earnings calendars, governance/accounting red flags, sensible sector balance.
  • Publish: candidates appear on the appropriate DeepList with their DeepScores.
  • Monitor & adapt: we track regime shifts and signal drift; updates reflect new information and data.

Suggested ways to use DeepLists

Mix and match based on your time horizon and risk tolerance.

Core

Core Trend Portfolio (lower maintenance)

  • Build from the 12M Trend list.
  • Hold 10–20 equal‑weight positions; review quarterly or semi‑annually.
  • Equal‑weight means ~5–10% per position when holding 20–10 names.
Tactical

Tactical Momentum Sleeve (more active)

  • Build from the 6M Momentum list.
  • 10–20 equal‑weight positions; review monthly or quarterly; refresh from the latest list.
  • Equal‑weight within this sleeve ≈ ~5–10% per name of the sleeve capital.
Satellite

High‑Conviction Mini‑Basket

  • 5–10 equal‑weight positions from the High‑Conviction list.
  • Review monthly; consider this a satellite to a core portfolio.
  • Equal‑weight here ≈ ~10–20% per position within the mini‑basket.
Blend

Blended approach

  • Split across horizons (e.g., 60% 12M Trend, 40% 6M Momentum) to balance persistence and responsiveness.
  • Equal‑weight the names within each sleeve; rebalance on your schedule.
  • Example: 60/40 with 10 names per sleeve → ~6% per name in Trend, ~4% per name in Momentum.

Practical tips

  • Prefer equal weights to avoid over‑confidence.
  • Diversify across sectors and, if relevant, across US and EU lists.
  • Rebalance on a schedule (don't chase every wiggle).
  • Consider simple exit rules that fit your risk tolerance (e.g., periodic refresh only, or use a long‑term moving‑average break as a discretionary stop). These rules are investor‑driven, not enforced by the lists.

What to expect

Momentum and trend strategies can rotate and will experience drawdowns. Plan for periodic rebalancing and benchmark‑relative variability.

  • Turnover: expect rotation as leadership changes.
  • Drawdowns: trend strategies can lag during sharp reversals or ranges.
  • False positives: probabilities are not certainties.
  • Benchmark awareness: measured against major indices.
Peak‑to‑trough swings are normal