Withdrawals, Rewards & the Validator Economy

The previous page, Slashing & the Inactivity Leak, showed the downside of staked capital: the protocol can burn it. That is only half the ledger of a validator’s life. This page closes the loop by tracing the money the other way — how staked ETH earns, and how it eventually comes home.

That second half was, for two and a half years, a promise the protocol had not yet kept. When The Merge went live in September 2022, staked ETH went in and could not come out: there was no code path to withdraw. Deposits were one-way. This was deliberate, and understanding why it was deliberate — and how the Shanghai/Capella upgrade of April 2023 finally opened the exit — tells you almost everything about how Ethereum balances a validator’s freedom to leave against the network’s need to stay secure. We end where the book’s throughline points: what does it cost to run one, and what do you earn for helping?

Where a validator’s income comes from

Recall from Validators & Staking 32 ETH that becoming a validator means locking 32 ETH. Nobody locks 32 ETH for nothing. In proof-of-work, the reward for securing the chain was the block subsidy (freshly minted coin) plus transaction fees, both paid to whichever miner won the hash race. Proof-of-stake keeps the idea — pay whoever does the work of securing the chain — but splits the income into four distinct streams, paid to different validators for different jobs.

   VALIDATOR INCOME (proof-of-stake)          replaces →   MINER INCOME (proof-of-work)
   ──────────────────────────────────────                 ────────────────────────────
   1. attestation rewards   (every epoch)                  block subsidy (minted coin)
   2. proposal rewards      (when chosen)      ┐
   3. priority tips         (when chosen)      ├─ paid to   transaction fees
   4. MEV                   (when chosen)      ┘  proposer

The four streams differ in one crucial way — who receives them and how often:

• Attestation rewards are the bread and butter. Every epoch, every active validator is expected to attest (vote) to the head of the chain, the correct source checkpoint, and the correct target checkpoint — the votes you met in Attestations & LMD-GHOST and Checkpoints, Casper FFG & Finality. A correct, timely attestation earns a small reward; a late or missing one earns less or a small penalty. This is a steady trickle every validator collects roughly once per epoch (~6.4 minutes), and it is the bulk of the base yield.
• Proposal rewards go to the one validator pseudo-randomly chosen to propose the block for a slot. A proposer packages other validators’ attestations into the block and is paid for including them. Because only one validator per slot proposes, and there are far more validators than slots, an individual validator proposes only occasionally — but the reward is larger when it happens.
• Priority tips are the sender-set portion of the EIP-1559 fee market. Recall from Ether as Fuel that a transaction’s fee has two parts: a base fee that is burned (destroyed, benefiting no one directly), and a priority tip the sender adds to jump the queue. The tip goes to the block proposer as ordinary income — the direct heir of the miner’s transaction-fee revenue.
• MEV (Maximal Extractable Value) is the value a proposer can capture by choosing how to order transactions in its block — front-running a trade, back-running an arbitrage, capturing a liquidation. This too accrues to whoever builds the block. In practice most proposers outsource block-building to specialists and receive the MEV as a bid, but the economic point stands: ordering power is worth money, and in PoS it is paid to the proposer instead of the miner.

The split matters because it separates a reliable, universal income (attestation rewards, which every honest validator earns) from a lumpy, occasional windfall (proposal rewards, tips, MEV, which only the chosen proposer earns). Long-run yield is dominated by the reliable stream; the lumpy streams add variance, not the baseline.

By the numbers — the two halves of a validator’s yield

Staking yield has two components with very different origins, and it helps to keep them separate:

   CONSENSUS-LAYER yield  (issuance)   depends on TOTAL ETH staked
      • attestation + proposal rewards, minted by the protocol
      • the more validators, the thinner each one's slice
      • roughly follows  reward ∝ 1 / sqrt(total staked)

   EXECUTION-LAYER yield  (fees + MEV)  depends on NETWORK DEMAND
      • priority tips + MEV, paid only to proposers
      • high in busy markets, near zero in quiet ones
      • independent of how many validators exist

As of 2024, the combined base yield for a solo validator has typically sat in the low-single-digit percent per year, with execution-layer income adding a variable top-up. Treat the exact percentage as time-sensitive — it moves with both the amount staked and network activity — but the shape is durable: one half shrinks as more people stake, the other half swells when the chain is busy.

Under the hood — the effective-balance cap at 32 ETH

Here is a subtlety that shapes everything downstream. Rewards do not scale with your actual balance; they scale with your effective balance, which the protocol caps at 32 ETH per validator.

Effective balance is a smoothed, rounded-down version of your real balance, used for two jobs: computing your rewards and penalties, and weighting your attestation in fork choice. It moves in whole-ETH steps (with hysteresis, so it does not flicker), and — the key rule — it never exceeds 32 ETH.

   actual balance:     32.0   33.7   40.0   35.2   (ETH)
   effective balance:  32     32     32     32     ← capped, never rises above 32
                        │
                        └─ rewards & vote weight are computed on THIS number

The consequence is stark: ETH sitting above 32 in a validator’s balance earns nothing. A validator with 40 ETH is rewarded exactly like one with 32; the extra 8 ETH is idle capital. This single rule is why partial withdrawals exist at all — without a way to move that excess out, rational stakers would either leave it earning zero or spin up new 32-ETH validators. It is also why, historically, large stakers ran many 32-ETH validators rather than a few fat ones. (A later upgrade, Pectra’s EIP-7251, raised this cap for validators that opt in — but the base model, and the one this page teaches, is the 32-ETH cap.)

Why withdrawals came after The Merge, not with it

Now the harder half. When The Merge shipped, you could deposit into staking but you could not withdraw. This was not an oversight — it was a staging decision, and it is a clean example of the book’s throughline: shipping a live re-architecting of a running system safely means changing one thing at a time.

The Merge’s job was already enormous: swap the consensus engine of a live chain securing tens of billions of dollars, without halting it (see The Merge). Bundling withdrawals into that same upgrade would have meant testing two intricate, irreversible mechanisms at once. If either had a bug, you could not tell which, and the blast radius touched everyone’s staked capital. So the roadmap split them:

   Dec 2020   Beacon Chain launches   → deposits open, PoS runs in parallel (no real duties yet)
   Sep 2022   The Merge               → PoS becomes canonical; execution + consensus fused
                                          ↑ staked ETH is now LOCKED — no exit path exists
   Apr 2023   Shanghai / Capella      → withdrawals enabled; the deposit becomes two-way

Shanghai and Capella are the same upgrade named on its two layers — Shanghai on the execution layer, Capella on the consensus layer — activated together in April 2023. (Ethereum upgrades are named this way because execution and consensus are separate clients that must be upgraded in lockstep; you will sometimes see the pair abbreviated Shapella.) Only after The Merge had run stably for over half a year, proving PoS could secure the chain, did the protocol take the second irreversible step and let staked capital flow back out.

There is a deeper reason the ordering was safe to defer, not just prudent. A validator whose stake is locked has every incentive to keep the chain healthy — their capital is trapped inside it. Making the stake withdrawable could only loosen that bond, so doing it later, once confidence was high, strictly reduced risk. The one-way door was a feature while trust was being earned.

Two kinds of withdrawal

Once Shanghai/Capella landed, a validator’s balance could leave the beacon chain in two very different ways. The distinction turns entirely on that 32-ETH effective-balance cap.

Partial withdrawal — skimming the excess

Because balance above 32 ETH earns nothing (the cap, above), the protocol automatically sweeps it out. A partial withdrawal — also called reward skimming — moves any balance above 32 ETH out of the validator and into a normal Ethereum account, while the validator keeps running with its full 32-ETH stake intact.

   validator balance: 32.9 ETH
                       ├── 32.0 ETH  → stays staked, keeps validating
                       └──  0.9 ETH  → swept out to the withdrawal address
   validator status:   STILL ACTIVE  (never left the set)

Three things make partial withdrawals special:

• They are automatic. There is no transaction to send and no queue to join. A background withdrawal sweep walks through validators in order, slot by slot, and any validator sitting above 32 ETH with a valid withdrawal address gets its excess skimmed. Over time every eligible validator is visited in turn.
• They require a withdrawal credential. A validator must have set an execution-layer (0x01) withdrawal address for the sweep to have somewhere to send the ETH. Validators created before this was standard had to perform a one-time credential change to become eligible.
• They do not touch the stake. The validator never leaves the active set, keeps its duties, and keeps earning. Skimming is just harvesting the interest without touching the principal.

Full withdrawal — a complete exit

A full withdrawal is a real departure. The validator exits the active set entirely: it stops attesting and proposing, and its entire balance — the 32-ETH principal plus any accrued rewards — is eventually swept out to the withdrawal address. This is how you get all your capital back.

   1. validator signs a VOLUNTARY EXIT message
   2. it joins the EXIT QUEUE  (rate-limited — see below)
   3. once exited, it stops all duties
   4. after a fixed delay, the withdrawal sweep pays out the FULL balance
   5. the 32 ETH (± rewards ∓ penalties) lands in the withdrawal address

Unlike a partial skim, a full withdrawal is initiated by the validator (a signed voluntary exit) and is not instant — it must pass through the exit queue. That queue is the last mechanism this page needs, and it is the one that protects everyone else.

The exit queue and the churn limit

Imagine full exits were instant and unlimited. A large fraction of validators could leave in a single epoch — whether from panic, a coordinated attack, or a mass migration to a competitor. The active validator set is what secures the chain: it is the set whose stake is at risk and whose votes finalize blocks (Checkpoints, Casper FFG & Finality). If that set could evaporate in minutes, an attacker could stake up, do damage, and flee before slashing or the inactivity leak could bite — and honest users could trigger a bank-run-style collapse of the security budget.

So the protocol rate-limits how fast validators can enter or leave, using a churn limit: a cap on how many validators may be activated or exited per epoch. Requests beyond the limit wait in a queue.

   voluntary exits this epoch:  ████████████████████  (many want out)
   churn limit (per epoch):     ████                  (only this many process)
                                └──► the rest wait in the EXIT QUEUE
                                     estimated wait = queue_length / churn_per_epoch

Two design details are worth internalizing:

• The churn limit scales with the size of the validator set (above a floor). More validators means a slightly larger absolute churn — but always a small fraction of the whole per epoch. The set can only turn over gradually, in either direction. (A later upgrade, Deneb’s EIP-7514, added a separate cap on the activation rate to slow how fast the set can grow, but the exit-side churn is the piece that matters here.)
• The queue is symmetric in spirit: entry is throttled so the set cannot balloon and dilute security guarantees faster than the network can absorb, and exit is throttled so it cannot collapse. Bounded churn in both directions keeps the active set a stable, known quantity from epoch to epoch — exactly what a security argument needs.

The exit queue is therefore not a nuisance; it is a security speed limit. It converts “everyone could leave at once” — a catastrophic possibility — into “the set changes slowly and predictably,” which is the property finality and slashing both quietly depend on.

It actually happened — the 2023 exit queue never broke

The real-world test of this design came right after Shanghai/Capella (April 2023). Commentators worried that enabling withdrawals for the first time would unleash a wave of exits and crash the ETH price as years of locked stake flooded out. It didn’t happen. The churn limit meant exits could only process a few thousand validators per day, so any wave was spread over weeks; and — the more telling result — far more of the early withdrawal volume was partial skims (validators harvesting rewards and staying active) than full exits. Enabling withdrawals made staking more attractive, not less: capital that had feared being trapped now had a proven door, so net stake kept climbing after the upgrade. The queue did exactly its job — turning a feared stampede into an orderly trickle — and the feared cliff never materialized. The lesson: a rate limit on leaving is what makes people comfortable enough to stay.

How it all fits: the long-run staking equilibrium

Put the pieces together and you can see the loop that governs participation. Rewards, penalties, the 32-ETH cap, and the churn limit are not four separate rules — they are one feedback system that steers how much ETH ends up staked.

   more validators join  ──► total staked ↑  ──► per-validator issuance yield ↓
        ▲                                                        │
        │                                                        ▼
   yield attractive?  ◄────────────────────────────────  yield now less attractive
        │                                                        │
        └──── if yes, more join (throttled by activation churn)  │
                                                                 ▼
                                       if no, some exit (throttled by exit churn)

Because issuance yield falls as more ETH is staked (reward ∝ 1/sqrt(total staked)), staking is self-balancing: high yield pulls capital in until yield falls; low yield lets capital drift out until yield recovers. The churn limits keep both movements gradual, so the equilibrium is approached smoothly rather than in destabilizing lurches. Penalties (for being offline) and slashing (for provable misbehaviour) tilt the incentive toward reliable participation specifically — you are paid to be present and honest, not merely to exist. And the 32-ETH cap ensures the security-relevant quantity (effective balance, and therefore vote weight) is uniform across validators, so no single validator’s failure or dishonesty is disproportionately weighted.

That is the validator economy in one sentence: a priced incentive to show up and vote honestly, a bounded punishment for failing, and a rate-limited door in both directions — tuned so untrusting strangers find it individually rational to keep the shared world computer running.

The architect’s lens

• Why does it exist? Because staked capital that can only go in is a trap no rational actor enters freely, and a security model needs both a reward for honest work and a credible, safe way to reclaim principal. Withdrawals and rewards together make staking a two-way, economically sane proposition instead of a one-way lock-up.
• What problem does it solve? It replaces the miner’s block subsidy and fees with a proof-of-stake income (attestations, proposals, tips, MEV), lets validators reclaim capital (partial skims and full exits), and — via the exit queue — prevents the active set from collapsing faster than the network can stay secure.
• What are the trade-offs? Exits are not instant: the churn limit can make you wait days or weeks to fully withdraw. That friction is the price of a stable security set. Rewards are also variable — issuance shrinks as more people stake, and fee/MEV income swings with demand — so staking yield is never a fixed guarantee.
• When should I avoid it? When you need your capital liquid on short notice, running a validator’s 32 ETH is the wrong home for it — the exit queue and the 32-ETH cap make it deliberately sticky. Liquid staking or simply not staking are the alternatives when instant access matters more than yield.
• What breaks if I remove it? Remove rewards and no one secures the chain — validating becomes pure cost. Remove withdrawals and staking is a one-way trap that suppresses participation. Remove the exit queue and the security set can evaporate in an epoch, letting attackers stake, strike, and flee before slashing can bite. Each piece holds up a different corner of the incentive structure.

Check your understanding

1. Proof-of-stake splits validator income into four streams. Name them, and explain which single stream makes up the bulk of a typical validator’s reliable base yield and why.
2. Withdrawals were deliberately staged after The Merge rather than shipped with it. Give the safety argument for that ordering, and name the upgrade (and its two-layer names) that finally enabled them.
3. Distinguish a partial withdrawal from a full withdrawal: what triggers each, what happens to the 32-ETH stake in each, and which one requires joining the exit queue?
4. Explain the role of the effective-balance cap at 32 ETH. Why does it mean ETH above 32 in a validator earns nothing, and how does that fact motivate the existence of partial withdrawals?
5. What is the churn limit, and what catastrophe does it prevent? Use the 2023 post-Shanghai experience to explain why a limit on leaving can actually make more people willing to stay.

Show answers

1. The four streams are attestation rewards, proposal rewards, priority tips, and MEV. Attestation rewards are the bulk of reliable base yield: every active validator is expected to attest each epoch, so this stream is earned universally and steadily (~once per epoch). The other three go only to the pseudo-randomly chosen proposer, so they are lumpy, occasional windfalls that add variance rather than baseline.
2. Bundling withdrawals into The Merge would have meant testing two intricate, irreversible mechanisms at once, so a bug could not be isolated and the blast radius covered everyone’s staked capital. Deferring withdrawals also tightened the incentive to keep the chain healthy (locked stake is trapped inside it), so opening the exit later — after PoS proved stable — strictly reduced risk. The upgrade was Shanghai (execution layer) / Capella (consensus layer), activated together in April 2023.
3. A partial withdrawal is triggered automatically by a background sweep whenever balance exceeds 32 ETH; it skims only the excess above 32, the validator keeps its full stake and stays active, and it does not use the exit queue. A full withdrawal is triggered by a validator-signed voluntary exit; the validator leaves the active set and its entire balance (principal plus accrued rewards) is swept out, and it must pass through the exit queue first.
4. Effective balance — a smoothed, whole-ETH value capped at 32 — is what rewards, penalties, and vote weight are computed on. Because it never exceeds 32, any actual balance above 32 contributes nothing to rewards, so that excess is idle capital earning zero. Partial withdrawals exist precisely to move that dead weight out automatically, so stakers are not forced to either waste it or spin up whole new validators.
5. The churn limit caps how many validators may activate or exit per epoch (scaling with set size above a floor), sending the overflow into a queue. It prevents the active validator set from collapsing (or ballooning) faster than the network can stay secure — e.g. an attacker staking, striking, and fleeing before slashing or the inactivity leak can respond. After Shanghai/Capella (April 2023) the feared stampede never came: exits were spread over weeks, most withdrawal volume was partial skims rather than full exits, and net stake kept rising — because a proven, orderly exit door made locking up capital feel safe rather than trapping.