Forecast cone
Range of likely prices over the next 3 months. Each card shows how often its 80% band actually held in past outcomes.
What is moving the 1-month odds
Approximate push of each factor, in percentage points of probability, after the model is shrunk toward the base rate by its tested skill.
Track record
Walk-forward test: every year was predicted only by a model trained on earlier years. "Base rate" means always predicting the historical share of up periods. A positive Brier skill means the model beat it. "Blend skill" tests the shrunk model the same honest way; unless it beats the base rate by more than 1 point, no tilt is published ("not used"). Band coverage tests the cone as published.
Calibration
When the raw model (before shrinking) said a given chance of going up, how often did prices actually go up? Points on the dashed diagonal mean the odds were honest; points flat around one height mean the model's odds carried no information. Bigger dots = more forecasts.
Live record
Every forecast the site publishes is logged on the day it is made and graded against the actual close once its horizon has passed. Unlike the backtest above, nothing here could have been tuned after the fact.
Download every logged forecast (CSV)
Recently graded forecasts
Dealer gamma (GEX)
Estimated from 15-minute delayed CBOE open interest, options expiring within 60 days. Positive gamma tends to dampen moves; negative gamma tends to amplify them. This is an estimate: actual dealer positions are not public.
Net GEX / 1%: dollars of stock dealers would trade for each 1% move. Gamma flip: the price where net gamma changes sign. Call / put wall: the strikes with the largest call-heavy and put-heavy gamma, which often act as resistance and support.
Macro backdrop
Presidential cycle
Unusual options activity
Contracts expiring in 5 to 60 days (same-week expiries always dwarf their open interest, so they are left out) whose volume today is at least 3× open interest and whose premium is at least $250K. Delayed data; this does not show which side of the trade was the buyer.
Economic calendar
Major US data releases and Fed decisions in the next two weeks, from the official schedules FRED publishes. Times are the usual release times (Eastern) and occasionally differ.
Earnings calendar
S&P 100 companies reporting in the next two weeks, with the consensus EPS estimate (via Alpha Vantage). Report dates can move: confirm with the company before acting on one.
Congress trade reports
Periodic Transaction Reports filed by House members under the STOCK Act, from the official House Clerk index. Each report lists one or more trades: open it for the tickers, dates and amount ranges. Members have up to 45 days to report, so the trades are already weeks old. Senate filings are not included.
Insider cluster buying
Open-market purchases by several insiders of the same company, from SEC Form 4 filings (via OpenInsider).
Market news
Methodology
Direction. For each ETF and horizon (1 week, 1 month, 3 months) a regularized logistic regression estimates the chance that the price ends higher. Inputs: 1, 3 and 12-month momentum, distance from the 200-day average, Williams %R, realized volatility, VIX and the VIX-minus-realized-vol premium, the 10y–2y curve, 3-month change in the 10-year yield, the Baa credit spread and its change, 3-month oil change, CPI inflation, presidential-cycle year and seasonality.
No look-ahead. Macro data is used only as of the day before. CPI is used only 50 days after its reference month starts. Each test year is predicted by a model trained on outcomes known before that year began. One limit remains: the backtest uses today's revised CPI history, not the values first published. Yields, spreads, oil and VIX are not revised.
Shrinkage. The published probability is the base rate plus the model's tilt, times a factor fitted on the walk-forward results. If the model showed no skill, the published odds are the base rate.
Cone. Forecasts the quoted price; dividends are not included. Width comes from the VIX, scaled by each ETF's volatility relative to SPY. The shape of each band (including how much more often markets fall hard than rally hard) comes from the actual distribution of past outcomes, measured walk-forward so each date only used outcomes already known. Coverage, including how often prices ended below or above the band, is tested on the cone exactly as published.
Not included yet. 13F institutional holdings.
For information and research only. Not investment advice. Markets are unpredictable and past patterns do not guarantee future results.