Datasets:

paperswithbacktest
/

Stocks-Weekly-PharmaClinicalPredict

pretty_name: Stocks Weekly PharmaClinicalPredict
language:
  - en
license: other
task_categories:
  - time-series-forecasting
  - tabular-regression
tags:
  - finance
  - quantitative-trading
  - backtesting
  - algorithmic-trading
  - stocks
  - equities
  - weekly
size_categories:
  - 10K<n<100K
extra_gated_prompt: >-
  This dataset is free to browse and gated for download. Approval is tied to a
  Papers With Backtest subscription, which also covers the other datasets in
  this organisation and the strategy catalogue at
  https://paperswithbacktest.com. Plans and what each one includes:
  https://paperswithbacktest.com/pricing
dataset_info:
  features:
    - name: trial_id
      dtype: string
    - name: source
      dtype: string
    - name: subsidiary
      dtype: string
    - name: sponsor
      dtype: string
    - name: official_title
      dtype: string
    - name: success_prediction
      dtype: float64
    - name: economic_effect
      dtype: float64
    - name: duration_prediction
      dtype: float64
    - name: success_composite
      dtype: float64
  splits:
    - name: train
      num_examples: 91981

Stocks Weekly PharmaClinicalPredict

Modelled outcome, duration and economic impact for pharmaceutical clinical trials, one row per trial.

91,981 rows, 9 columns. Updated by Papers With Backtest.

Why It Matters

A biotech's value is a probability-weighted pipeline, and the probabilities are what this dataset estimates:

Event odds ahead of the readout: success_prediction is a modelled probability that a trial reaches its endpoint. Priced against the sponsor's market capitalisation, it is a view on how much of the outcome is already in the stock.
Timing: duration_prediction estimates how long the trial still has to run, which is what decides whether an event is inside or outside a holding period.
Size of the move: economic_effect scores how material the trial is to the sponsor, separating a phase III lead asset from a minor indication.

Load It

Installation/Upgrade:

pip install --upgrade pwb-toolbox

Load the Dataset:

from pwb_toolbox import datasets as pwb_ds

predictions = pwb_ds.load_dataset("Stocks-Weekly-PharmaClinicalPredict")
trials = pwb_ds.load_dataset("Stocks-Weekly-PharmaClinicalTrials")

# The predictions carry no ticker; the trial register is what maps them to one.
linked = predictions.merge(
    trials[["trial_id", "symbol", "datetime"]].drop_duplicates("trial_id"),
    on="trial_id",
    how="inner",
)

Columns

Column Name	Description
trial_id	Registry identifier for the trial. The join key to `Stocks-Weekly-PharmaClinicalTrials`.
source	Registry the trial was read from.
sponsor	Sponsoring organisation.
subsidiary	Subsidiary running the trial, where it differs from the sponsor.
official_title	Registered title of the trial.
success_prediction	Modelled probability, between 0 and 1, that the trial meets its endpoint.
success_composite	Composite score combining the success model with trial characteristics.
duration_prediction	Modelled remaining duration, in days.
economic_effect	Unitless index of how material the outcome is to the sponsor.

What This Data Does Not Cover

There is no ticker and no date column. The file is one row per trial, keyed by trial_id, and it carries neither symbol nor datetime. Mapping a prediction to a listed company means joining Stocks-Weekly-PharmaClinicalTrials, which holds both. Anything that needs a point-in-time view has to take its timestamps from that side too.

Access

Browsing the card and the schema is open to anyone. Downloading the files needs an approved request, tied to a subscription: what each plan includes. The same subscription covers the other datasets in this organisation.

Elsewhere

Dataset page and coverage charts
The strategy catalogue, 3,806 papers and 4,837 replicated strategies
pwb-toolbox, the loader used in the snippet above
awesome-systematic-trading, the replicated strategies with their measured Sharpe
Every dataset in this organisation

Papers With Backtest publishes 32 datasets on the Hub and codes the papers that use them. Every strategy in the catalogue is run over its own full history before it is published, which is where the numbers above come from.