Predicting the success of new crypto-tokens: the Pump.fun case

TL;DR

This work investigates how early on-chain trading dynamics on Pump.fun predict the graduation of newly launched tokens in a bonding-curve launchpad. By conditioning graduation probability on the bonding-curve state $vSol$ and on behavioral signals (e.g., liquidity velocity, bot activity, top traders, and creator activity), the authors reveal that rapid liquidity accumulation and a higher share of non-bot trades strongly correlate with successful migrations to the real AMM, while bot-dominated activity tends to hinder graduation. They introduce an economic breakeven curve to translate probabilistic forecasts into an intuitive profitability benchmark and document widespread pump-and-dump behavior, including pre-graduation liquidations that help explain the limited long-run value of many launches. Overall, the paper provides a scalable, on-chain predictive framework for early-stage token traction in bonding-curve platforms and highlights design-induced incentives that shape pre- and post-graduation market dynamics.

Abstract

We study the dynamics of token launched on Pump.fun, a Solana-based launchpad platform, to identify the determinants of the token success. Pump.fun employs a bonding curve mechanism to bootstrap initial liquidity possibly leading to graduation to the on-chain market, which can be seen as a token success. We build predictive models of the probability of graduation conditional on the current amount of Solana locked in the bonding curve and a set of explanatory variables that capture structural and behavioral aspects of the launch process. Conditioning the graduation probability on these variables significantly improves its predictive power, providing insights into early-stage market behavior, speculative and manipulative dynamics, and the informational efficiency of bonding-curve-based token launches.

Figures (13)

• Figure 1: On-chain trading activity in the PumpFun ecosystem during September 2025. The top panel reports the daily number of newly created tokens. The second panel shows the daily number of trades. The third panel displays the average daily price of Solana (SOL) in USD. The bottom panel reports the daily trading volume, expressed in million USD, computed from SOL-denominated volumes rescaled by the corresponding daily average SOL price.
• Figure 2: Cumulative signed amount value of Solana (vSol) circulating through all newly created tokens during September 2025, including both bonding curve and post-graduation AMM activity.
• Figure 3: Empirical survival curve of bonding-curve progress. For each threshold of SOL in bonding curve, we report the number of newly created tokens reaching that amount. The ordinate is shown on a logarithmic scale, highlighting the rapid drop-off in the fraction of tokens that accumulate large amounts of SOL along the bonding curve.
• Figure 4: Graduation dynamics of Pump.fun tokens across multiple dimensions. Top left: Empirical distribution of the number of bonding-curve steps required to reach graduation. Top right: Empirical distribution of time to graduation (in minutes) from token creation. Both distributions are strongly right-skewed, with medians of approximately $457$ steps and $4.4$ minutes, respectively (dashed lines). Bottom left: Normalized fraction of tokens whose maximum bonding-curve state exceeds a given value of vSol, showing that only a small fraction of launched tokens reaches the graduation region and migrates to the real PumpSwap AMM pool. Bottom right: Joint distribution of graduation steps and time to graduation, with color intensity indicating the number of tokens on a logarithmic scale. The concentration of mass at short times and low step counts highlights the fast and momentum-driven nature of successful launches, while the sparse tail reflects heterogeneous and less efficient paths to graduation.
• Figure 5: Empirical probability of graduation as a function of the virtual Solana threshold (vSol). The solid line ($p_{\mathrm{std}}$) reports the baseline estimator $P(\mathrm{grad}\mid \exists\,x:x>vSol)$. The dashed line ($p_{\mathrm{std}}^{t>2}$) shows the same probability after excluding ultra-fast graduations with time to graduation shorter than two seconds. Light-blue bars ($p_{\mathrm{bin},j}$) display a non-parametric binned estimate over intervals of width $5$ SOL. The close agreement between the curves indicates that extremely rapid, likely bot-driven events do not materially affect the graduation-probability profile, while the monotonic increase reflects the deterministic progression of tokens along the bonding curve toward the graduation threshold.