Methodology
What EdgePulse does, how, and what it cannot do.
Data
Daily closing prices of front-month futures from Yahoo Finance, one source and one unit per market. Nothing is mixed in from other providers, because switching sources mid-series creates fake jumps.
| Market | Category | Contract | Unit |
|---|---|---|---|
| Brent Oil | Energy | BZ=F | $ / barrel |
| Natural Gas | Energy | NG=F | $ / MMBtu |
| Gold | Safe haven | GC=F | $ / troy oz |
| USD Index | Safe haven | DX-Y.NYB | index points |
| Wheat | Food | ZW=F | $ / bushel |
| Corn | Food | ZC=F | $ / bushel |
| Rough Rice | Food | ZR=F | $ / cwt |
Wheat and corn trade in cents per bushel and are shown in dollars. A price is dropped only when it jumps sharply for a single day and immediately reverses, which is almost always a data error. Front-month futures roll to the next contract periodically, which can add small artificial moves.
The 90% one-month range
EdgePulse does not publish a price target. For commodities, a simple "price stays the same" guess is very hard to beat, and the previous version of this site proved it by losing to exactly that.
Instead it answers a question that matters for budgets and contracts: how far could this price realistically move in the next 30 days?
- Current volatility. An exponentially weighted average of recent daily price changes (decay 0.94), so recent turbulence counts more than calm months a year ago.
- The market's own history. For every past day, the 30-day move that followed is divided by that day's volatility. Only moves whose outcome was already known on the forecast date are used.
- The range. The 5th and 95th percentiles of those standardized moves, scaled by today's volatility, give the low and high prices. Fat tails and skew in each market are captured from its own data instead of assuming a bell curve.
If the method is well calibrated, about 90% of outcomes land inside the range. The track record shows how close it gets.
The trend signal
"Uptrend" means today's price is above where it was 63 trading days ago (about three months). Strength compares that move with normal volatility: weak, moderate or strong. It describes where the price has been. As a predictor of the next month it has not beaten always guessing "up" in the backtest, and the track record shows that openly.
The track record therefore compares it with two baselines: a coin flip (50%) and always predicting "up". A trend signal that cannot beat "always up" adds nothing.
Scoring
- Live forecasts are made every weekday from that day's close and stored immediately, starting 2026-10-16.
- Backtests run the identical code on past data, one forecast a week per market, using only prices available on each date.
- A forecast is scored against the first closing price on or after its due date. The scoring happens in the database, so it cannot be edited after the fact.
Limitations
- Volatility estimates react with a lag. Ranges are often too narrow right after a sudden shock, such as a war or supply disruption.
- The trend signal is a weak indicator, not a prediction. Do not base a trade or a contract on it alone.
- Futures prices are not the prices a business pays. Freight, grade, basis and currency differences can be large.
- Yahoo Finance data is free and unofficial. Gaps and occasional errors are possible.
EdgePulse is a portfolio project. It is for information only and is not financial or trading advice.