How We Detect GE Dumps Before They Happen

May 26, 2026 · 8 min read · Technical

Every OSRS flipper has been there. You check the Grand Exchange, buy an item at what looks like a fair price, and watch it crash 30% over the next hour. By the time you realize what happened, you're holding a bag.

Grand Flip Out detects these dumps in real time. Not after they've happened. Not with a 5-minute delay. In real time, scored from 0 to 100 based on severity, with recovery timeline estimates.

Here's how the math works.

The Problem: GE Prices Don't Crash Randomly

A dump isn't just "the price went down." Prices fluctuate constantly. What makes a dump different is sustained negative drift — the price keeps falling, doesn't bounce back, and volume patterns shift in a specific way.

Regular price noise looks random. A dump has structure:

The question is: how do you detect the dump at Day 0, not Day 7?

Layer 1: CUSUM (Cumulative Sum Control)

CUSUM is a statistical process control technique from manufacturing quality assurance. It detects when a process shifts away from its expected mean — exactly what happens during a dump.

How it works: We track the cumulative sum of price deviations from the recent average. When this sum crosses a threshold, it means the price has shifted to a new regime — not just random fluctuation, but a genuine trend change.

For each item, we maintain two CUSUM statistics:

When S_low exceeds our threshold, we flag the item as "dumping." The threshold is adaptive — volatile items need a higher threshold to avoid false alarms, while stable items trigger on smaller deviations.

CUSUM catches dumps that simple percentage-change alerts miss, because it accounts for the sustained nature of the drift. A 5% drop that bounces back won't trigger CUSUM. A 3% drop that continues for 6 consecutive polls will.

Layer 2: Z-Score Anomaly Detection

Every 5 minutes, we compute the z-score for each item — how many standard deviations the current price is from its recent mean. This gives us a universal "how unusual is this?" metric across all 4,400+ items.

Z-score alone isn't enough (items can naturally be volatile), but combined with CUSUM regime detection, it tells us both that something unusual is happening AND how severe it is.

Layer 3: VPIN (Volume-Synchronized Probability of Informed Trading)

VPIN measures order flow toxicity. In the OSRS context, it estimates whether the current trading activity is driven by informed traders (who know something you don't) or uninformed traders (regular flippers).

High VPIN on an item means someone is aggressively selling with information — maybe they know about an upcoming nerf, or they're dumping a position ahead of bad news. GFO tracks VPIN across 1,966 items and flags 20+ elevated items at any given time.

Layer 4: Dump Scoring (0-100)

When multiple layers agree, we compute a composite dump score:

Scores above 65 are "confirmed dumps." Scores 40-65 are "likely dumps." Below 40 is "watch but don't act."

Layer 5: Recovery Prediction

Detecting the dump is only half the value. The real question is: should you buy it?

We use an Ornstein-Uhlenbeck mean-reversion model to estimate recovery probability and timeline. This is the same model used in quantitative finance to price mean-reverting assets.

The key parameter is theta (θ) — the speed of mean reversion. Higher theta means faster recovery. We estimate theta from recent price history and combine it with volume data to predict:

Layer 6: Neural Network Confidence

Grand Flip Out is the only OSRS tool with a neural network. Our living-nn system trains 24/7 on GE data, learning patterns that statistical methods can't capture.

The NN doesn't replace the statistical layers — it provides a confidence modifier. When the NN agrees with the dump score, confidence goes up. When it disagrees, we flag uncertainty. This multi-layered approach means we catch dumps that any single method would miss.

Layer 7: Recipe Ripple Chains

This is unique to GFO. When one item dumps, related items often follow — but with a delay.

Example: Nature rune crashes 20%. Within 1-3 hours, alchable items see a demand surge (alchemy becomes more profitable). If you detect the nature rune dump at Hour 0, you can position in alchables before Hour 1.

We track 16+ production chains (ore→bar→item→alch, herb→potion→consumption, etc.) and predict cascade effects in real time.

What This Means for You

When GFO detects a dump, you get:

All of this is free. No paywall. No premium tier for dump alerts. The intelligence is the product.

See dump alerts in real time

Open Dashboard

Or join Discord for instant dump notifications

The Competition

Other tools show you that a price dropped. We show you why it dropped, whether it will recover, and when to buy. That's the difference between information and intelligence.

07Flip uses basic percentage-change alerts. GE Tracker shows historical charts. Neither uses CUSUM, VPIN, mean-reversion modeling, neural networks, or ripple chain analysis. We use all of them, simultaneously, on every item in the GE.

Try it at grandflipout.com. It's free.

Grand Flip Out by Granary Works. Information-only. Jagex-compliant.

Privacy · Terms · Discord