With an annual income of hundreds of millions and aggressive buybacks, why is Pump.fun still being "shorted" by the market?

Original Author: Four Pillars

Original Compilation: AididiaoJP, Foresight News

Key Points

In eight months, 99.5% of daily protocol revenue was used to complete a $328 million buyback, with two independent data aggregators reaching the same conclusion despite not communicating with each other. To manipulate the data, one would need to deceive both DeFiLlama and Adam_tech, which only indexes Solana's Dune data, maintaining a stable ratio of 68-69%, and have 105.17 billion PUMP in verifiable wallets as support.

The "dilution curve" in August 2026 is actually a supply replacement rather than an addition; at the current revenue level, the buyback can absorb twice the new supply. When the team and investors unlock their tokens, community emissions will cease. Monthly emissions will drop from 10 billion to 9.2 billion.

The real reason for the current valuation multiple being suppressed lies in: industry classification ("original sin stock" nature), trust foundation (anonymous team, discretionary buybacks), and capital flow (suspected insider traders using buybacks to sell off).

1. $328 Million Buyback Evidence

Rumors of revenue fraud at Pump.fun have circulated on Twitter, and the following analysis indicates that these rumors are unfounded.

As of March 15, 2026, data from fees.pump.fun shows that the cumulative buyback amount reached $328 million. This means using 2,283,518 SOL to purchase 10.45 billion PUMP, accounting for 10.45% of the total supply, offsetting 29.52% of the circulating supply. Over eight months, the daily buyback amount maintained between 99.5% and 100.5% of protocol revenue, with an average daily buyback of $1.25 million as of February 2026. Revenue fraud would require substantial funding: for every dollar bought back, one dollar of SOL must flow out from a verifiable wallet to purchase tokens stored in an auditable address. To fabricate $328 million in revenue, an actual expenditure of $328 million is necessary.

The relevant tokens are stored on-chain and can be verified (as of March 17, wallet G8CcfRff holds 10.396 billion PUMP, and 8PSmqJy6 holds 1.21 billion PUMP, totaling 10.517 billion). The initial execution wallet 3vkpy5Y (marked as "Pump Buy Back" by Solscan) completed transfers to holding wallets and has been rotated out as of August 2025, with a current balance of zero.

DeFiLlama recorded a total protocol revenue of $300,041,880 from July 15, 2025, to February 21, 2026. During the same period, the cumulative buyback amount from fees.pump.fun was $300,178,162. The matching degree between the two is 100.05%, with a discrepancy of only $136,000 between the two independent systems.

Adam_tech's Dune dashboard provides a third layer of verification. This platform only tracks revenue on the Solana chain, consistently accounting for 68-69% of DeFiLlama's multi-chain data, as it did not index the Padre revenue launched on Base, Ethereum, and BNB Chain in October 2025. This ratio remains stable daily, indicating that both independently read the same on-chain events.

Before the launch of PumpSwap in March 2025, the error between the three data sources was within 1-5%. After the launch of PumpSwap, the data diverged into three layers: total fees, protocol revenue, and Solana-only revenue. If revenue data were artificially fabricated, it would require deceiving two independent on-chain indexers simultaneously, maintaining a stable cross-ratio during three product changes, keeping the multi-chain revenue split ratio consistent with actual business expansion, and supporting it with tokens purchased from verifiable wallets.

2. Four Statistical Tests

In addition to on-chain evidence, 747 days of fee data can undergo four standard tests to verify the authenticity of financial data. While individual tests may not be conclusive, the credibility significantly increases when all four tests point to the same conclusion.

The first test examines the fee-to-revenue ratio, which is the hardest metric to fabricate. Pump.fun collects fees from each joint curve transaction, but not all are counted as protocol revenue; some flow to LPs, creators, and referral rewards. In the dataset, the total fees to net revenue ratio dropped from 1.0 to about 0.48, but not in a gradual manner; instead, it fell sharply in three phases, each corresponding to product changes recorded on-chain:

• On March 20, 2025, the LP fee split mechanism was launched with PumpSwap, causing the ratio to drop from 1.00 to 0.70 within two days.
• On May 13, the creator revenue sharing mechanism was launched, dropping the ratio from 0.69 to 0.56.
• On September 2-3, the Ascend project introduced a dynamic fee mechanism, with tiered pricing allowing creators to earn up to 0.95% fees on low market cap tokens, leaving only 0.05% for the protocol, reducing the ratio from 0.68 to 0.46.

Fabricating this data would require simultaneously simulating the fee and revenue sequences through three structural adjustments, with daily ratios fluctuating between 0.40 and 0.55 based on token tier combinations. This complexity makes fabrication difficult. The reality is that product iterations naturally lead to data changes, rather than artificially constructing structural breakpoints that align with contract deployment times.

The second test examines continuity and numerical distribution characteristics, aiming to determine whether the data exhibits signs of human input. Humans struggle to generate truly random sequences, tend to avoid long streaks, prefer integers, and unconsciously favor specific numbers. The data from Pump.fun does not exhibit these characteristics:

The longest consecutive increase or decrease was 6 days, with an average consecutive length of 1.92 days, consistent with expectations of a natural process with moderate momentum. The distribution of consecutive lengths decreases geometrically: 185 instances of single-day consecutives, 111 instances of two-day consecutives, 52 instances of three-day consecutives, down to 7 instances of six-day consecutives.

The last digit of daily fees is nearly uniformly distributed between 0-9, with each digit accounting for 8.7%-11.2%. 88.8% of the dates have non-integer last digits, with only 7 days among 743 non-zero dates ending in 00 or 000.

The third test examines the weekend effect. Pump.fun is a retail platform, with users primarily issuing tokens on trading days rather than weekends. The average weekday fees are $2.14 million, while weekend fees are $1.81 million, showing a consistent decline of about 18%, evident week after week over two years of data. The Mann-Whitney test shows a p-value of 0.003, indicating statistical significance. If data were artificially constructed, one would need to deliberately keep weekend figures consistently low, increasing the complexity of the fraud and the risk of being uncovered.

The final test examines autocorrelation, measuring the correlation between today's revenue and tomorrow's revenue. Pump.fun's first-order lag autocorrelation is 0.78, indicating a 78% correlation between today's fees and yesterday's; after a week (lagged 7 days), it remains at 0.65; after two weeks (lagged 14 days), it is 0.57. This slow, smooth decay reflects the momentum characteristics of organic platform activity: active periods cluster, while sluggish periods persist. If daily revenue were randomly generated, the correlation between adjacent dates would be nearly zero, and the data would jump like noise rather than flow like a market. Fabricating a single lag's high autocorrelation is not difficult, but simultaneously fabricating the entire decay structure (with each lag gradually decreasing monotonically) while maintaining the weekend effect, continuity characteristics, and true numerical distribution is nearly impossible.

Four independent tests, four consistent conclusions, and three data sources corroborate each other. The revenue data is authentic and credible.

3. Analysis of Remaining Valuation Discount Factors

Rumors of revenue fraud are one of the reasons for the current valuation pressure on PUMP. The previous analysis has clarified this point. However, the token is still trading at a discount, necessitating an exploration of other suppressive factors and their authenticity.

First, analyze the team unlock in August. Community emissions of 10 billion tokens per month will cease after reaching 240 billion in July, coinciding with the team and investor unlock, totaling 9.2 billion tokens per month. The monthly emission volume drops from 10 billion to 9.2 billion, reducing the inflation rate by 8%. Based on the current average daily revenue of $1.25 million, the monthly average buyback is $38 million, which can absorb about twice the new monthly supply of approximately $19 million. After August, as emissions decrease and buybacks continue, this ratio will further improve.

Revenue has also not shown a decline. Over the past fourteen months, monthly fees have fluctuated between $2.3 million and $4.8 million daily: a 49% drop in July 2025, a 94% rebound in August, a 72% surge in September, and a 45% spike in January 2026. Overall, it reverts around a daily average of $2.5 million, with weekly trading volumes stabilizing at $640-700 million. The so-called "decline from Q3 to Q1" is a one-sided conclusion drawn from selectively picking peak data from September.

Remaining suppressive factors are as follows:

The "original sin stock" discount is the most persistent. Solidus Labs found that 98.6% of tokens on the platform exhibit "rug pull" characteristics, which creates an expected effect: regardless of revenue, institutional allocators will not include "meme coin casinos" in their portfolios. This is a continuous structural factor, entirely unrelated to revenue quality.

Source: Solidus Labs

Suspected insider selling constitutes real recent pressure. Wallet 77DsB received 3.75 billion PUMP from an address marked as "token custody wallet" by Solscan in July 2025, reportedly liquidating for 8.02 million USDC between February 16 and 22, 2026. Wallet GpCfm simultaneously transferred 1.21 billion PUMP ($2.57 million) to Bitget. A third wallet deposited 1.757 billion PUMP ($3.54 million) into Bitget on March 6. Although there are no sources confirming actual ownership, at least $14 million in funds flowed to exchanges at a market price of $0.002 during the protocol buyback within thirty days, while the previous private placement price was $0.004. Regardless of the wallet owner's identity, this situation raises suspicions.

Trust is the hardest aspect to price. The founder is anonymous (co-founder Dylan has a "rug pull" record from 2017); the buyback is explicitly "discretionary" ("pump.fun may modify or terminate related plans at any time"); Bubblemaps once indicated that Hayden Davis was associated with a $50 million private placement, which was later deleted after co-founder Alon claimed it was "slander." On-chain associations do exist, but ownership is disputed and unverified.

The above factors do not pertain to the business fundamentals. Revenue is real, data supports it, and the unlock arrangements are favorable to holders. The "original sin stock" label, anonymous founders, and insider capital flows all contribute to a trust discount for a protocol with an average verifiable on-chain revenue of $1.25 million and a buyback capable of absorbing twice the new supply. The trust discount will eventually narrow; this scale of revenue will not remain mismatched permanently.