DGPT Standings Forecast — a Monte Carlo model

A Monte Carlo simulation that forecasts the end-of-season standings of the Disc Golf Pro Tour (DGPT) — i.e. each pro's probability of finishing 1st, top-8, top-16, etc. across the MPO and FPO divisions — rather than just predicting a single outcome.

Inspired by FiveThirtyEight's sports forecasts. It began as a Google Sheets model (capped near ~50 simulations by spreadsheet recalculation time) and was ported to Python/pandas to run 1,000+ full-season simulations.

How it works

1. Inputs (events/, eventplayers/, pointslogic/):
   • the season's event schedule — date, tour, event type, number of rounds;
   • each event's projected field, with every player's PDGA rating;
   • the DGPT points tables by tour and finishing position.
2. Rating → expected score. Each player's PDGA rating becomes an expected strokes-vs-field-average per round (≈ 6 rating points per stroke), with a time-decay term so events further in the future carry more uncertainty.
3. Simulate (N = 1,000). For every player at every event, draw each round from a normal distribution (σ ≈ 6.8 strokes), sum to an event score, rank the field, and resolve first-place ties with a random playoff.
4. Score it. Apply DGPT points by tour/finish, split points across ties, and apply each series' "best N of M events count" rule (Elite top 8, Silver top 3, NT top 4, PDPT top 4). Completed events use real results in place of simulated.
5. Aggregate. Across all 1,000 runs, compute each player's standings distribution — P(1st), P(top 8 / 16 / 32) and average finish — per division and tour. Results land in results/.

Repository layout

DGPTModelV2.ipynb    the current model (load → simulate → score → aggregate)
events/              event schedules (date, tour, type, rounds)
eventplayers/        projected fields + PDGA ratings, per scenario
pointslogic/         DGPT points tables (Elite / Silver / NT / PDPT)
results/             aggregated forecast outputs (MPO & FPO)
other/               earlier model version + a results-scraping experiment

Techniques shown

Monte Carlo simulation · probabilistic forecasting · pandas data wrangling · rating-based score modeling · sports analytics.

Status & caveats

• Built for the 2021 DGPT season (data snapshots dated Aug 2021). It's a point-in-time exploratory notebook, not a maintained library — to run a different season you'd swap in that season's CSVs.
• DGPTModelV2.ipynb is current; other/DGPTModel.ipynb is the earlier version and other/ScreenscrapeTest.ipynb an experiment for scraping live results.
• The final aggregation cell is intentionally verbose (one block per finishing position) — function over polish.
• Player ratings, schedules and results are public/factual PDGA data.

License

MIT.