Prediction market spread modelling and analysis
Laen...
Kuupäev
Autorid
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Tartu Ülikool
Abstrakt
Likviidsuse tagamine ennustusturgudel on saanud vähem tähelepanu kui traditsioonilistel finantsturgudel. Käesolevas magistritöös uuritakse, kas Polymarketil vaadeldud hinnavahed ja täitumisintensiivsuse mustrid järgivad logitteisendatud Avellaneda–Stoikovi turutegemise raamistiku ennustusi. Logitteisendust kasutatakse seetõttu, et ennustusturgude hinnad on piiratud nulli ja ühe vahele. Analüüsis kasutatakse Bitcoini-strike turu andmeid 2026. aasta märtsist. Andmetest hinnatakse volatiilsus ja täitumisintensiivsuse kahanemisparameetrid. Riski parameetrit, ajahorisonti ja inventuuri hoitakse fikseerituna. Mudeli põhjal arvutatud hinnavahet võrreldakse seejärel Polymarketil vaadeldud hinnavahega. Tulemused toetavad raamistikku vaid osaliselt.
Ostupoole täitumisintensiivsused enamasti vähenevad hinnapakkumise kauguse kasvades, nagu mudel ennustab, kuid sobivus on päevade ja hinnavahemike lõikes nõrk ja ebastabiilne. Müügipoole hinnangud on aga sageli vale märgiga: täitumisintensiivsus suureneb hinnapakkumise kauguse kasvades, mitte ei vähene. Päevadel, mil nii ostu- kui müügipoole kahanemisparameeter on positiivne, on mudeli põhjal arvutatud mediaanhinnavahe 1.8 kuni 5.4 korda suurem vaadeldud hinnavahest.
Liquidity provision in prediction markets has received less attention than in traditional financial markets. This thesis tests whether observed bid–ask spreads and fill-intensity patterns on Polymarket follow the predictions of a logit-transformed Avellaneda–Stoikov market-making framework. The logit transform is used because prediction-market prices are bounded between zero and one. The analysis uses tick-level order-book and trade data from a single high-volume Bitcoin-strike market in March 2026. Belief volatility and fill-intensity decay parameters are estimated from the data. Risk aversion, the quoting horizon, and inventory are held fixed. The model-implied spread is then compared with the observed Polymarket bid–ask spread. The results give only partial support for the framework. Buy-side fill intensities mostly decrease with quote distance, as the model predicts, but the fit is weak and unstable across days and price ranges. Sell-side estimates often have the wrong sign: fill intensity rises with quote distance rather than falling. On days where both buy- and sell-side decay estimates are positive, the modelimplied median spread is 1.8 to 5.4 times the observed bid–ask spread. The framework is therefore a theoretical benchmark, not a direct description of observed order-book behaviour in this Polymarket sample.
Liquidity provision in prediction markets has received less attention than in traditional financial markets. This thesis tests whether observed bid–ask spreads and fill-intensity patterns on Polymarket follow the predictions of a logit-transformed Avellaneda–Stoikov market-making framework. The logit transform is used because prediction-market prices are bounded between zero and one. The analysis uses tick-level order-book and trade data from a single high-volume Bitcoin-strike market in March 2026. Belief volatility and fill-intensity decay parameters are estimated from the data. Risk aversion, the quoting horizon, and inventory are held fixed. The model-implied spread is then compared with the observed Polymarket bid–ask spread. The results give only partial support for the framework. Buy-side fill intensities mostly decrease with quote distance, as the model predicts, but the fit is weak and unstable across days and price ranges. Sell-side estimates often have the wrong sign: fill intensity rises with quote distance rather than falling. On days where both buy- and sell-side decay estimates are positive, the modelimplied median spread is 1.8 to 5.4 times the observed bid–ask spread. The framework is therefore a theoretical benchmark, not a direct description of observed order-book behaviour in this Polymarket sample.
Kirjeldus
Märksõnad
prediction market, market making, modelling, ennustusturg, turutegija, modelleerimine