LoginSign Up
← Back to all sparks
P

PINstimation

INFRA · APIS
Velocity0.0

Estimation of the Probability of Informed Trading

github.com/monty-se/pinstimation ↗

A market-microstructure toolkit that keeps adding estimators as the papers land.

market-microstructurefinanceinformed-tradingr-packagehigh-frequency-dataestimation
◆Current state
PINstimation estimates probability-of-informed-trading models — PIN, multilayer PIN, adjusted PIN and VPIN — from trade and quote data, and handles the trade classification and aggregation that feeds them. The current 0.2.0 adds ivpin(), a maximum-likelihood variant of VPIN from Ke and Lin (2017). The package's early history is compressed into a single hour of backfilled tags in October 2022, so version order there does not track release order.
◆Where it's heading
Each release tracks the literature: a Bayesian PIN estimator from Griffin et al., an improved VPIN from Ke and Lin, initial-parameter generation realigned to Ersan and Ghachem. The other steady thread is data handling — matrix inputs so the estimators compose with rolling windows, user-specified aggregation frequencies, and now quote leads as well as lags. The three-year gap between 0.1.2 and 0.2.0 makes this a slow, publication-paced package rather than an actively developed one.
◆Prediction
On this pattern the next release adds whatever estimator the authors publish next, since two of the three feature releases here implement a specific paper. Nothing in the entries points to a change in the package's structure.

◆Recent moves

  1. 8mo ago

    PINstimation 0.2.0

    Adds ivpin(), a maximum-likelihood reworking of VPIN that the notes say holds up better with small volume buckets or infrequent informed trades — the conditions where the original estimator is least stable. Trade classification also accepts negative time lags, treating quotes as leads rather than lags.

    View source ↗
  2. 3y ago

    PINstimation v0.1.2

    classify_trades() exposes per-trade buyer or seller initiation directly rather than only inside the aggregation step, and aggregate_trades() accepts a frequency instead of always producing daily data. Together they open intraday work that the daily-only pipeline had ruled out. Also corrects mpin_ecm(), which could admit information layers of zero probability.

    View source ↗
  3. 3y ago

    PINstimation v0.1.1

    The estimators accept matrix input rather than dataframes only, which is what lets them run inside rolling-window helpers like zoo's rollapply. Adds pin_bayes() for Bayesian PIN estimation and fixes calculation bugs in vpin() and negative trade intensities in initials_pin_ea().

    View source ↗
  4. 3y ago

    PINstimation v0.0.1-beta

    A beta tag stamped one minute after the initial release and describing the same pin_bayes() addition and vpin() fixes that 0.1.1 carries. Part of the backfilled October 2022 history rather than a distinct release.

    View source ↗
  5. 3y ago

    PINstimation v0.1.0

    Labelled the initial release, with a two-line body noting a future::plan reset fix. It sits inside the same hour as the 0.0.1-beta and 0.1.1 tags, so this is release history being backfilled rather than the package first appearing.

    View source ↗