Research foundations
Academic sources These references inform the breakout evidence and quantitative research adaptations shown by Casper. The site uses its own screening rules; published returns and classification accuracy do not establish its trade success rate.
Carlo Zarattini, Andrea Barbon, Andrew Aziz · 2024 · Working paper
Informs intraday logic: Freeze the opening range; keep catalyst evidence separate; compare volume with an appropriate baseline.
How Casper differs: Casper uses a 30-minute range, two completed five-minute closes, and estimated VWAP. This differs from the paper’s main five-minute stop-entry strategy. Recent-bar volume is not its same-time 14-session relative volume.
Study limitations and implementation notes Supplied PDF pp. 9, 13–15, 19
The principal backtest is a five-minute opening-range stop-entry strategy with directional opening-candle rules; Casper uses a 30-minute range, two completed five-minute closes and estimated VWAP. The study uses approximately 7,000 US stocks over 2016–2023 and a point-in-time universe including delisted stocks. Casper's selected AI-related watchlist is a different universe. The paper includes a per-share commission assumption; this does not establish Casper's attainable spread, slippage, stop-fill quality or borrow availability. The bulk screener does not have a matched 14-session opening baseline. Selected-ticker month history can supply 14 valid prior 09:30–09:35 ET intervals. The helper excludes invalid/duplicate sessions, records baseline dates and refuses fewer than 14. Its five-minute comparison is separate from Casper's 30-minute setup and same-session recent-bar volume. The supplied page 14 denominator prints t−1 within a sum indexed by i; the surrounding text and page 19 describe prior-14-day opening-interval comparison. We do not claim an exact formula replication. Reported historical returns and ex-post best/worst ticker tables are not probabilities, current buy/sell recommendations or proof of forward performance. Md. Siam Ansary · 2022 · Journal article
Informs swing logic: Keep prior-range consolidation evidence distinct from a later observed break.
How Casper differs: The paper uses two-week closing-price bounds and a 98% minimum/maximum consolidation test. Casper uses a prior 20-session high and five-session range/volume contraction. The reported classifier metrics are not a calibrated trade forecast.
Study limitations and implementation notes Journal pp. 53–55; supplied PDF pp. 2–4
The paper describes 5,630 initial Yahoo samples from 2000–2022, under-sampled to 432 balanced rows, then an 80/20 training/test split. A chronological holdout, walk-forward test, ticker-level holdout and purging of overlapping windows are not documented in the supplied methodology. That limits inference about future trading performance; leakage is not proven solely by this omission. The label is constructed from consolidation and an observed closing-price breakout. Classification accuracy can describe recognition of that label rather than future return prediction. Standardization is described before the split, without a documented train-only fitting procedure. Leakage-safe preprocessing cannot be verified from the supplied report. Two-week closing bounds and minimum close >98% of maximum close differ from Casper's prior 20-session high and five-session true-range/volume comparisons. No transaction-cost trading backtest, calibrated success probability or earnings/macro validation is provided for Casper's universe. Lu Liu, Jianrong Wei, Jiping Huang · 2013 · Journal article
Informs intraday and swing logic: Show failure levels and benchmark context separately from upside confirmation.
How Casper differs: The paper’s centered extrema window includes later observations. Its retrospective R/R′ measures are not used as live signals or probabilities. Casper only uses prior bars and already completed observations.
Study limitations and implementation notes Supplied PDF pp. 2, 4–5
Local maxima are defined on [t−Δt, t+Δt]. Future observations and a later extremum are required for the paper's R/R′ measurements; using them at t would introduce look-ahead. The analysis covers 500 S&P 500 constituent stocks and 847 Shanghai stocks through July 31, 2012, with multiple window lengths. This is a distribution study, not a current AI-stock strategy backtest. A fitted distribution or exponent is not the probability that today's specific breakout succeeds. Casper does not calculate R/R′ or assume identical upside/downside behavior across companies and horizons. Quantitative research adaptations Citations and research descriptions supplied by the site owner; full-text verification pending.
Gil Cohen · 2025, as supplied
Application: Keep multi-target review levels separate from input evidence. Casper displays one- and two-ATR reference levels around the last completed close.
Implementation limits: These ATR levels are Casper's own descriptive scenarios, not the paper's target predictions. XGBoost, LightGBM, AdaBoost, stacking, blending and transfer learning are not trained or deployed here.
Glucksman & Lahanis · NYU Stern, 2025, as supplied
Application: Separate momentum-following candidates from pullback/recovery candidates and compare completed stock returns with SPY on identical dates.
Implementation limits: Casper screens individual stocks; it does not reproduce online portfolio-selection weight updates or the paper's backtests. Lookbacks and thresholds are Casper defaults. No historical profitability is claimed.
arXiv:2404.00424v1 · 2024, as supplied
Application: Inspect medium and longer price windows together: 20- and 60-session returns and excess returns, plus a 50-session trend check.
Implementation limits: Feature agreement is deterministic. It is not a Transformer or transfer learning. Historical sentiment labels, trained weights and an independent chronological holdout are unavailable, so a sentiment-based return forecast is unavailable.
Rule definitions
Screening methodology Casper narrows your watchlist using transparent price and volume rules. Every match can be traced to the criteria you selected. Setup reasoning separates observed evidence, confirmation, and failure levels. Thresholds are research defaults; they have not been validated as success probabilities.
Academic sources behind the setup logic Near breakout Within 2% below the highest high of the previous 20 daily bars, a prior five-completed-session base range of 5% or less, and above the prior 50-completed-session simple moving average. Reading a swing setup The prior five completed sessions define the base. We compare their mean true range and volume with the preceding 15 sessions, and the completed-bar 50-session average with its value five sessions earlier. Contraction is descriptive evidence; it does not prove accumulation. A separate confirmation rule requires a completed daily close above the prior 20-session high and full-session volume ≥1.5× the previous 20-session average. An unfinished daily bar never confirms this rule. Opening range breakout Freeze the high and low of six completed five-minute bars from 9:30–10:00 ET. Require two consecutive completed five-minute closes above that high in the current session and price above estimated session VWAP. Missing opening bars and imported intervals other than five minutes cannot qualify. The reasoning panel compares the latest five-minute volume with up to 20 preceding bars in that session; this is not same-time relative volume and does not gate this preset. A later close back inside the range flags a failed hold. Intraday momentum Daily change of at least 1% from the previous daily close and price above estimated regular-session VWAP. Missing opening bars or gaps make the VWAP comparison unavailable. The estimate weights each five-minute bar's typical price (high + low + close) / 3 by its volume. Volume expansion Today's cumulative volume is at least 1.5 times the average of the previous 20 complete daily bars, with a nonnegative daily change. This is a full-day volume ratio, not a same-time relative volume measure. Daily research & options Research views are screening heuristics, recalculated from observed price history and options context. Above a rising prior 50-session average with a positive daily move gives a bullish technical view; the inverse gives a bearish view. Opposing recent option activity makes the combined view mixed. Put/call volume ≤0.6 is called call-heavy and ≥1.2 put-heavy, with at least 100 reported contracts at strikes within ±15% of provider spot across the sampled expirations and a latest trade in the expected regular session. These thresholds are product choices, not validated probabilities or signed trade flow. Options do not independently establish a trade direction. Capital positioning The 20-session money-flow proxy weights the closing position within each daily high-low range by volume, divided by total volume. Values above 0.05 or below −0.05 are descriptive accumulation or distribution flags. They do not identify the trader. Institutional holder rows preserve each report date; provider changes may refer to different reporting periods. SEC 13F filings are quarterly and generally due within 45 days of quarter-end. Earnings readiness Company KPI playbooks link to issuer reporting. Yahoo financials are provider-standardized actuals with fiscal period, units and accounting basis. Estimated earnings dates require issuer confirmation. Business KPI actuals, guidance and consensus targets are not invented when the feed lacks them. Funds show fund-specific watchpoints rather than corporate earnings. Extended hours Available completed five-minute observations from 04:00–20:00 ET can be shown on intraday charts. Shading separates premarket and after-hours bars. Extended quotes have their own timestamp and reference price; regular-session screening, opening-range levels and estimated VWAP remain based on regular hours. Data & coverage Missing indicators fail active filters. Current Yahoo intraday screens require a recent completed bar from the active regular session, no more than 30 minutes behind. Older data remains inspectable under All tracked names, with age labels. This age ceiling is a product rule, not a provider guarantee. Market holidays and early closes follow the published NYSE calendar for 2026–2028; unsupported dates do not qualify for current-session screens. Delayed prices show the bar close timestamp separately from the fetch time. Imported files stay in this browser tab. Sample prices are invented to demonstrate the interface and are not market observations. Educational references: Fidelity on breakout confirmation and Schwab on volume . Casper’s exact thresholds are our screening choices.
Provider documentation · Market calendar Data connection
Data sources and refresh Connecting to Yahoo Finance. Prices will appear when the connection is ready.
Yahoo prices. No API key. Prices and history load automatically from Yahoo Finance through yfinance. The screener uses daily history and completed five-minute bars. Intraday charts can include extended hours, while regular-session setup rules retain their regular-hours inputs. Options, holders and earnings load for the selected ticker. Linked headlines refresh separately every minute; company results retain their original report dates. Each stock shows its latest bar timestamp; prices may be delayed.
Price snapshots are cached for 45 seconds, with 15-second edge caching and revalidation. Completed daily baselines are reused for up to 15 minutes; observed regular five-minute bars update the unfinished daily candle. A partial aggregate never confirms a daily breakout. A recent fetch does not guarantee a recent price; each stock shows its own data age. Missing history never passes an active filter. On first load, unavailable data produces a retry screen. A failed refresh keeps the last loaded prices and labels the issue.
Use sample data yfinance is an unofficial feed intended for personal research. See the data-source documentation for usage terms.
Import your price history Upload a CSV with symbol, time, open, high, low, close, volume, and timeframe (1D, 5m, or 15m). Include 50 completed daily sessions plus the observed day for moving-average screens. Files stay in this tab and are never uploaded.
Choose CSV file Download example format Paste CSV instead Price-history CSV Import pasted history 
