About metrics

How the metrics on the site work and how to read them

Where the data comes from

We get the prices and the list of leagues from poe.ninja. Every few minutes the site saves a “snapshot” of the prices of a league, and all the figures on the page are built from these snapshots

A league does not get into the metrics automatically: it has to be added separately. Leagues without metrics are not shown on the site

The prices in the metrics are expressed in a certain currency, and every currency has its own set of metrics. The list of currencies comes from the base currencies of poe.ninja, and some currencies for the metrics can be added separately, for example Mirror of Kalandra. You can pick the currency in the header of the league page

The item basket

A day is divided into several slices of a few hours each (the borders of the slices are in UTC). We call such a slice together with the data collected in it a bucket. The price snapshots of the whole league over this period go into the bucket, and one median price is taken for every item from them. The median keeps a random spike in a single snapshot from distorting the picture

Rules for getting into the basket:

An item is in the basket if its price is not below the threshold. The threshold is its own for every currency: this way cheap and noisy items do not affect the estimate of the market
If an item was in the basket last time, it stays there even if it got cheaper than the threshold. Otherwise it would drop out of the calculation exactly when its price moves a lot
If an item has no price in the bucket at all, it drops out of the basket
The number of trades does not affect getting into the basket: the most expensive items are traded rarely, and a selection by turnover would almost empty the basket. A low turnover is taken into account later, when the return is calculated

Market metrics

We estimate the market by the movement of the items of the basket. Three numbers are calculated for every bucket

Market return: How much the prices changed on average compared with the previous bucket. The median of the changes of all items is taken, not the mean, so that a few sharp spikes do not skew the result. Items with few trades count with a lower weight or do not count at all. A value of 0.05 roughly means that prices rose by 5%
Items in the calculation: How many items actually took part in the calculation of the market return
Items in the basket: How many items there are in the basket in total, including those that have no price change or too few trades. The larger the share of the items in the calculation, the more reliable the estimate of the market. If it is small, the estimate can be very “noisy”

A sharp market return means a “market day”: almost everything moved at once, for example a jump in the price of the currency itself. On such days the conclusions about single items are unreliable

Item metrics

For every item of the basket we calculate the following

Price: The median price of the item in the bucket, in the selected currency
Price change / Δ price: How much the price changed compared with the previous bucket. It is calculated through a logarithm, but for small moves it is simply a fraction: 0.05 roughly means +5%, −0.05 roughly −5%. If the item had no price in the previous bucket, there is no figure
Trades per hour / Trades/h: How many units of the item were sold per hour on average. The fewer the trades, the easier it is for a random trade to move the price, and the less we trust the price change. For items we do not trust fully, the value is shown in gray
Trust in the price change: A number from 0 to 1. If there are fewer than 2 trades per hour, we do not take the price change into account at all. From 20 trades per hour we trust it fully, and between these values the trust grows gradually. Items without trust do not take part in the calculation of the market and of the beta
Residual: The main figure. It is the part of the price change that the movement of the market does not explain: the change of the market multiplied by the beta of the item is subtracted from the price change of the item. It shows whether the item moves on its own rather than with everyone else. The residual has to be compared with the usual fluctuation of this item, not with one percent for all
Usual fluctuation of the residual / Fluctuation: How far the residual of the item usually deviates from zero over the last 7 days. It is calculated so that single outliers hardly affect it. If the residual is clearly larger than the usual fluctuation (a rule of thumb: twice as large), the move is considered unusual and gets into the corresponding list
Residual relative to the fluctuation / Multiple: How many times the residual of the item is larger than its usual fluctuation. A value of 2 or higher means an unusual move, the sign shows the direction: plus means the price rose more than the market explains, minus means it fell more
Number of observations / Sample: How many values were used to estimate the usual fluctuation. The window is designed for 14 buckets (7 days): the closer to 14, the more reliable the estimate. If there are few observations, the item is marked “low data”

Beta and R²

These two figures are calculated from the history of the item and are recalculated once a day from the data of the last 7 days. The old values are not overwritten, so the page of the item shows how the beta changed

Beta (β): Shows how strongly the item reacts to the movement of the market. β = 1: the item moves together with the market. β = 1.5: when the market rises by 10%, the item rises by about 15%. β close to 0: the item hardly follows the market. While there is little data, the beta is pulled towards 1 so that a short history does not give a random number
R²: What part of the fluctuations of the item is explained by the movement of the market, from 0 to 1. A high value: the item follows the market and its residual is small. A value close to 0: the item lives on its own and has to be judged by its own movement

How to use this

These are guidelines, not ready trading signals

Look at the residual, not at the plain price change: this separates the own movement of an item from the general one
Compare the residual with the usual fluctuation of this very item. For common items the typical deviation is about 2%, for rare ones it is several times larger
Confirm a shift with several buckets in a row (2–3)
Trust the conclusions less on “market days” and when the market calculation has few items
Do not rely on estimates marked “low data”
For items with R² close to zero the market explains almost nothing, so judge them only by their own fluctuation