Ethereum DEX Trades

One Parquet table with 54 columns: one row per swap our indexer decodes on Ethereum DEXes, with both sides, the pool, fees, the transaction and a USD value. Failed trades stay in with a flag, and NFT marketplace trades stay in too.

About 618 million trades over 123 months, roughly 175 GB as Parquet, estimated from the sample files. The last 24 months hold about 243 million rows, the last 12 about 132 million and the latest month about 12 million. Row counts come from the live table; exact figures arrive with the delivery manifest.

To the start of our index: from 9 July 2016, at block 1,854,431, the first swap in our index, to yesterday.

Uniswap v2, v3 and v4 carry most rows; the last month also has Balancer, PancakeSwap v3, Fluid, Curve, Bancor, 1inch and the Seaport NFT marketplace, 27 protocol names in that month. The protocol name describes the pool design, so group by Trade_Dex_OwnerAddress, the factory, to split forks that share a design.

The export applies no filter of its own: failed trades, spam tokens and zero-value trades all stay in, and the columns let you drop them. NFT sales on Seaport are kept: 242 of the 6,364 sample rows, and 21% of trades on a sampled day in June 2022. Filter the two Fungible columns to drop them.

Each swap takes its USD value from one side, the safer one: a stablecoin (USDT, USDC, DAI and BUSD) first, then ETH or WETH, then any other priced token, at the block's hourly price from our own hourly Ethereum price series, which has covered about 8,300 tokens over its life. The same value goes on both sides, so a thin token cannot inflate a swap against ETH. On one sampled day a year, between 92% and 99.9% of trades had a stablecoin, ETH or WETH on one side; the free sample has a USD value on 98.4% of rows. A value of 0 means no price for either side.

Decoded from Ethereum by our own indexer, which reads each DEX's swap events and ties them to the transaction and the calling contract. USD values are added at export from our price series. Before release we ran the export query over blocks 26,047,176 to 26,047,275 and read the two sample files back: 6,364 trades, 54 columns, no empty values, USD on 98.4% of rows. Rerun over the blocks of our older 2025 sample file, it found every row in that file, plus 17 the live table gained since.

For one token, filter the contract column on either side to its address and sum Trade_Buy_AmountInUSD over successful swaps. For a v2 or v3 pool, filter Trade_Dex_SmartContract to the pool; for Uniswap v4 and PancakeSwap Infinity, filter Trade_PoolId, since their pools share one contract.

A routed swap through several pools is several rows: 65% of sample trades share a transaction with another swap, so sum by transaction for trader volume and by pool for DEX volume. The buyer address is usually a router; Transaction_From is the trader. Failed trades are 9.6% of the sample, some inside successful transactions, so filter Trade_Success. The 2016 rows are a handful a day, 8 on a sampled December day; a sampled day in June 2017 had 793 trades and one in June 2018 had 18,454.

One line in DuckDB: SELECT * FROM read_parquet('dex_trades/*.parquet') LIMIT 10. Pandas works too: pandas.read_parquet(path). The sample file opens without credentials.

Refreshed daily with T+1 latency, so a purchase made today includes everything to yesterday. Delivered as Parquet named by block range, 50 blocks per file and sorted ascending, under a stable S3 layout with a JSON manifest listing every file and its sha256. Signed HTTPS links are emailed once the files are prepared, or on request we deliver into your own S3, GCS or R2 bucket.