Crypto Cost Basis Missing Records: How to Reconstruct What the IRS Expects

The single biggest reason crypto taxpayers end up with IRS problems that are worse than they should be is missing cost basis records. The exchange went under, an old wallet was lost, the CSV export never got saved, the ledger was on a laptop that died. The IRS position is clean and unforgiving: if you can't prove basis, basis is zero. That turns a legitimate 40% gain into a 100% gain. This is a straight walk-through of the rules, and five practical methods to reconstruct crypto cost basis missing records.

The burden: IRC §1012 and who has to prove what

Under IRC §1012, the basis of property is its cost. The burden to substantiate basis is on the taxpayer — not the IRS. In Burnet v. Houston, 283 U.S. 223 (1931), the Supreme Court held that failure to prove basis results in basis of zero. That rule has been applied to crypto in Tax Court memoranda and IRS audit practice.

Translation: if the IRS audits a 2021 crypto sale and you don't have records, the IRS starts at zero basis and counts the entire proceeds as gain. You have to affirmatively prove otherwise.

The Cohan rule: the taxpayer's only lever when records are gone

In Cohan v. Commissioner, 39 F.2d 540 (2d Cir. 1930), Judge Learned Hand held that where a taxpayer has clearly incurred a deductible expense but cannot substantiate the precise amount, the Tax Court should make an estimate bearing heavily against the taxpayer whose inexactitude is his own fault. The Cohan rule is alive and applied — in Estate of Baker v. Commissioner, in Williams v. Commissioner, and in dozens of Tax Court memoranda on crypto and stock basis reconstruction.

The Cohan rule is not a free pass. It requires:

• Credible evidence that the expense (or basis) was actually incurred.
• Some reasonable basis for estimating the amount.
• An estimate that bears heavily against the party with the missing records — meaning conservative, IRS-favorable assumptions on close calls.

§274(d) restrictions on strict substantiation (meals, entertainment, travel, listed property) do not apply to cost basis under §1012. Cohan estimation is available for crypto basis reconstruction.

Five reconstruction methods

1. Exchange CSV exports (even from dead exchanges)

Every major exchange offers a transaction history export — Coinbase, Kraken, Gemini, Binance.US, Crypto.com. Pull every export you can. For dead exchanges (FTX, Celsius, Voyager, BlockFi, Cred, Mt. Gox), the bankruptcy trustees have released partial data, and some exchanges return transaction history on request even post-bankruptcy. Chainnotice, Koinly, and CoinTracker all maintain integrations with dead-exchange data exports where available.

2. Bank and card records

Every fiat on-ramp left a bank record. Pull three to seven years of bank and credit card statements. Match debits to exchanges (Coinbase ACH, Kraken wire, etc). For each fiat deposit into an exchange, you have a dollar-denominated basis event. Combined with contemporaneous exchange pricing (historical data is free from CoinGecko, CoinMarketCap, Kaiko, or CCData), you can often reconstruct basis on a first-in-first-out basis from nothing but bank statements.

3. Blockchain analytics

Services like Chainalysis, TRM, Elliptic (enterprise), or consumer tools like Blockchair, Etherscan, and Solscan, let you walk a wallet's complete transaction history. For every inbound transfer, you have a timestamp and an amount. Cross-reference the timestamp to historical price data and you have a valuation. On-chain data is permanent. Even if you lost every exchange record, the blockchain still has the receipts.

This is particularly powerful for self-custody wallets. Connect the wallet to Koinly or CoinTracker, let it pull the chain data, and reconcile.

4. Email and cloud records

Search Gmail, iCloud, and any old inbox for "Coinbase," "Kraken," "Binance," "confirmation," "receipt," "deposit." Every exchange emails confirmations. A single email with a transaction ID and timestamp supports basis. Also search for wallet-backup emails, KYC confirmations, and ACH notifications.

5. FIFO vs specific identification — pick the right method

The default lot identification method for crypto is FIFO (first-in-first-out) absent adequate identification under Treas. Reg. §1.1012-1(j). Specific identification requires, as of 2024 final regs and the 1099-DA rollout, contemporaneous records identifying the specific lot sold.

In a reconstruction, FIFO is almost always friendlier because it doesn't require per-lot evidence. Sometimes that produces higher gain than specific identification would have; in reconstruction cases, pick the method that you can defend with the records you actually have.

Worked example: reconstructing a 2021 ETH position

Facts: taxpayer bought ETH repeatedly on Coinbase and Gemini from 2017-2021, moved some to a Ledger, used some on Uniswap. Sold $430,000 worth in late 2021 across multiple exchanges. Never tracked basis. Received CP2000 in 2024 proposing zero basis — $430,000 of reported gain.

1. Coinbase full transaction history: pulled 2017-2021. Every fiat purchase has a date, USD amount, and ETH amount. Cost basis from Coinbase: $142,800.
2. Gemini history: pulled via their legacy export tool. Cost basis: $38,100.
3. Ledger wallet: imported into Koinly. The ETH moved in from Coinbase and Gemini already has basis from steps 1-2. One inbound transfer from an unidentified wallet (1.2 ETH) — no basis traceable. Apply Cohan: estimate basis at FMV on receipt date ($1,840) with 20% haircut = $1,472.
4. Uniswap activity: three swaps. Disposition events with their own basis (traced back to Coinbase purchases), and three new token positions with basis equal to the ETH given up.
5. Total reconstructed cost basis: $189,500.

Result: $430,000 proceeds − $189,500 basis = $240,500 of actual gain, not $430,000. Tax saved at the 23.8% long-term cap gains and NIIT combined rate: about $57,200.

The IRS accepted the reconstruction. The key was the contemporaneous documentation for every traced dollar — plus a narrow Cohan claim on the one unidentified inbound that had supporting blockchain evidence.

The zero-basis worst case

If you cannot produce any credible evidence of basis, the IRS will tax the full proceeds as gain. On a $430,000 sale, that's roughly $102,000 in federal tax at long-term rates plus NIIT — on a position that may have had $240,000 of legitimate gain, not $430,000. The difference between a reconstructed case and a zero-basis case is often 50-70% of the tax. That's what this work is worth.

The 1099-DA transition: the rules just tightened

For 2025 calendar-year transactions, centralized exchanges issue Form 1099-DA reporting gross proceeds. For 2026 transactions, they will also report cost basis. Once basis reporting is live:

• The exchange's basis becomes the presumptive basis on the taxpayer's return unless the taxpayer reports otherwise on Form 8949 with Code B.
• The AUR matching program will flag any mismatch between the 1099-DA basis and the reported basis.
• For pre-2025 positions held on exchange and sold in 2026+, the exchange may report "basis not available," which flags zero basis to the IRS system.

The window to get ahead of basis reconstruction is now. Clean up pre-2025 positions, document the basis, and move it to the exchange (where possible) or retain the Form 8949 supporting documentation for when the 1099-DA mismatch comes.

Documentation standards we use

For every reconstructed case I handle, I build a basis workpaper with: source document (exchange CSV, bank statement, email, blockchain transaction ID) for every acquisition; historical price source and methodology for any FMV estimates; a dated narrative explaining any Cohan-based estimates and the IRS-favorable adjustments applied. That workpaper is retained with the tax file and is what we hand to an examiner if the return is challenged. No shortcuts.