What is the ETF arbitrage mechanism that keeps price close to NAV?

It is a primary-market create/redeem loop run by authorized participants who can swap an ETF’s underlying basket for ETF shares, or swap shares back for the basket. When the ETF is rich, they create shares and sell them, increasing supply. When it is cheap, they redeem shares, shrinking supply, and the gap typically narrows.

Who can do NAV arbitrage in ETFs?

Only an authorized participant can create and redeem ETF shares directly with the issuer. Other traders can attempt secondary-market pair trades between the ETF and its holdings, but they cannot access the conversion step that mechanically changes ETF share supply.

Why do ETFs still trade at a premium or discount if arbitrage exists?

Arbitrage is not frictionless, so APs act only when the premium or discount is wide enough to cover spreads, execution costs, and hedging risk. Small deviations and bid-ask spreads can persist because the basis is not large enough to clear that all-in cost hurdle.

Is iNAV the same as NAV for spotting mispricing?

No. iNAV is an intraday estimate that many ETFs publish frequently, around every 15 seconds, but methodologies vary and it can be unreliable when underlying markets are closed or illiquid. A premium/discount can also be a timestamp mismatch between last price and the NAV reference.

Why is bond ETF arbitrage weaker than equity ETF arbitrage?

BIS explains that bond ETF creation and redemption baskets can differ substantially from actual holdings, with overlap as low as less than 3% for the largest bond ETF. Because the conversion is less direct and the underlying bonds are less liquid, price/NAV gaps can widen and persist, especially in stress.

How ETF arbitrage keeps price near NAV

ETF price stays close to NAV because authorized participants can swap a basket of underlying assets for ETF shares, or swap ETF shares back for the basket, in large creation units. That convertibility creates a two-market basis trade that tightens big gaps, but only when the premium or discount is wide enough to pay for spreads, hedging, and execution risk.

Key Takeaways

ETF price tracks NAV because authorized participants can create shares when the ETF is rich and redeem shares when it is cheap, changing share supply until the gap narrows.
The ETF arbitrage mechanism is a two-market basis trade with a cost hurdle, so small premiums/discounts and bid-ask spreads can persist.
Intraday iNAV updates frequently, but it can be stale or model-driven when underlying markets are closed or illiquid, making some “nav premium” readings optical.
Bond ETF arbitrage can be structurally weaker because creation/redemption baskets may barely overlap with holdings, with BIS citing less than 3% overlap for the largest bond ETF.

Why ETF prices can drift from NAV

NAV is an accounting number built from the portfolio: the aggregate value of holdings minus liabilities, expressed per share. The exchange price is a live auction outcome. Those two numbers are not computed from the same inputs at the same time, so a gap is not a scandal. It is the normal result of an exchange-traded instrument meeting a portfolio that is valued on a schedule.

The drift shows up as a premium or discount. A premium means the ETF’s market price is above NAV, and a discount means it is below. Traders often describe that gap as a nav premium, but the important detail is what is being compared: a last traded price on an exchange versus a NAV that may be based on closing prices (equities), bid or mid marks (bonds), or a fixing (commodities), with fees sitting on the liability side. Invesco’s institutional note makes the timing problem concrete with a time-zone mismatch example: a London-listed Nasdaq 100 UCITS ETF can print a “last price” at the London close while the NAV is computed off the US close. That comparison can manufacture a premium or discount that is not a clean arbitrage signal.

This is where the broader creation/redemption plumbing matters. ETFs have a secondary market where anyone can trade shares intraday, and a primary market where share supply can expand or contract. That primary market is the core of how etf creation and redemption works, and it is the reason ETFs usually do not behave like closed-end funds that can wander far from asset value for long periods.

How authorized participants link markets

The key actor is the authorized participant, a large institution with the contractual right to create and redeem ETF shares directly with the issuer. Positioned’s glossary is explicit on exclusivity: only APs can do primary-market create/redeem, which makes them the bridge between the issuer and the exchange. Everyone else is confined to the secondary market, trading existing shares.

That exclusivity is the entire edge. Without it, “ETF arbitrage” would just mean trying to replicate the portfolio and hoping the spread converges. With it, the AP can convert one leg into the other in size, at rules-based terms, by delivering a basket to the issuer or receiving a basket from the issuer. The unit of trade is not one share. It is a creation unit, which Positioned describes as typically 25,000–50,000 shares. That scale matters because it turns a small per-share premium into a meaningful dollar P&L, and it concentrates the activity in firms that can finance, hedge, and clear the underlying basket.

APs are not doing charity work for ETF holders. They run a basis book. When the ETF is “rich” versus the basket, the book wants to be short the ETF and long the basket, then flatten via creation. When the ETF is “cheap,” the book wants to be long the ETF and short the basket, then flatten via redemption. That is the authorized participant arbitrage loop in its simplest form.

The constraint is that APs are not obligated to step in. Positioned flags AP withdrawal risk during stress, and BIS describes March–April 2020 as a period when bond ETF price/NAV gaps were not fully arbitraged away. The linkage exists, but it is conditional on the trade being executable.

The creation and redemption arbitrage loop

The mechanical force that pulls price back toward NAV is share supply control. Creations add ETF shares when the ETF trades at a premium, and redemptions remove shares when it trades at a discount. That is the part most explanations wave at, but it is the only part that actually changes the object being traded on the exchange.

When the ETF is at a premium, the loop is:

1. The AP buys the underlying basket in the open market. This is the “cheap leg” if the ETF is trading rich. 2. The AP delivers that basket to the issuer and receives newly created ETF shares in a creation unit, typically 25,000–50,000 shares. 3. The AP sells those ETF shares on the exchange. The added share supply pressures the ETF price down toward NAV.

When the ETF is at a discount, the loop flips:

1. The AP buys ETF shares on the exchange. This is the “cheap leg” when the ETF is trading below the basket value. 2. The AP redeems those shares with the issuer and receives the underlying basket. 3. The AP sells the underlying basket. The removal of ETF shares from circulation pressures the ETF price up toward NAV.

This is the ETF premium arbitrage story, but with the missing clause that makes it real: the AP only presses create/redeem when the basis clears an all-in cost hurdle. Positioned and Investopedia both emphasize that arbitrage is not frictionless. The hurdle includes bid-ask spreads in both legs, market impact in the basket, hedging slippage while the trade is being built, and financing and settlement constraints. That is why “near NAV” does not mean “at NAV,” even in very liquid products.

The other structural detail is that creation/redemption is typically done in-kind, meaning securities are swapped for securities rather than cash. Positioned highlights in-kind as a defining feature. For the arbitrage loop, in-kind matters because it keeps the conversion between ETF shares and the basket operational without forcing the fund itself to transact for cash at the worst moment.

How NAV and iNAV are observed intraday

Most screens show at least three different “values” for an ETF: last price, NAV, and iNAV. Confusing them is how traders talk themselves into phantom mispricings.

NAV is the portfolio-based per-share value, calculated from the holdings and liabilities. It is not a live executable price. Invesco points out that underlying assets are valued differently depending on what they are, and that the last traded price is simply where the exchange cleared supply and demand. Comparing those two without checking timestamps is how a clean-looking premium/discount turns into an apples-to-pears comparison.

iNAV is the intraday estimate meant to update frequently. Investopedia and Invesco both describe iNAV updates around every 15 seconds. The catch is that iNAV is produced by an iNAV provider and methodologies vary. Invesco warns that iNAV can be unreliable when the underlying market is closed or illiquid, because the calculation may be forced to use stale prints or proxies. The London-listed versus US-close example is the simplest version of the problem, but the same issue shows up any time the ETF trades in one session while a meaningful slice of the basket is not actively pricing.

For traders and users, the sanity check is chronological, not philosophical. When a premium/discount looks wide, the first question is whether the ETF’s last price and the NAV reference are built off the same market moment. If they are not aligned, the “gap” may not be tradable NAV arbitrage at all. It may be a measurement artifact.

This is also why the arbitrage mechanism is best thought of as two markets with a conversion option, not as NAV acting like gravity. The conversion is real, but the reference point used to label a premium or discount can be noisy.

When arbitrage breaks down or weakens

The tether is strongest when the underlying basket is liquid, continuously priced, and easy to source in size. Equity ETFs tend to fit that template because the creation/redemption basket is usually close to the actual holdings, and the underlying equities trade on exchanges with transparent prints. BIS uses that contrast to set up where the textbook story breaks first: bond ETFs.

BIS’s March 2021 Quarterly Review explains that bond ETF arbitrage operates differently because the baskets used for creation and redemption can differ substantially from the ETF’s actual holdings. The striking data point is overlap: BIS notes that for the largest bond ETF, the basket included less than 3% of the bonds in the actual holdings. That design choice is not an error. It reflects bond-market microstructure, including lower liquidity, larger minimum trade sizes, and the fact that many bonds trade over the counter. The consequence is that “convert ETF to holdings” is less literal. The AP is converting ETF shares into a negotiated basket that is meant to be representative or operationally feasible, not necessarily identical to the portfolio.

That structural decoupling weakens the clean NAV arbitrage story. It can also change what a discount means during stress. BIS documents that in March–April 2020, bond ETF prices deviated strongly from NAV and the gap was not fully arbitraged away by APs. Positioned’s glossary adds the behavioral reason: APs are not obligated to create or redeem on any given day, and they can step back when pricing the underlying becomes too risky.

Historical tape shows the same pattern outside bonds. Investopedia cites SEC reporting that during the 2010 flash crash, 27% of 838 ETPs were temporarily “unhinged” from their underlying securities pricing. That is the clean reminder that the arbitrage loop is conditional. When execution is impaired, the conversion trade that normally pins price toward NAV can become slow, expensive, or simply not worth running.

Near the end of the chain, it circles back to creation/redemption plumbing. The more direct and liquid the conversion described in how etf creation and redemption works, the tighter the expected band around NAV. The more proxy-like the basket and the harder the hedge, the more the market should expect the band to widen.

The Take

I’ve watched people treat NAV like a “true price” and then get confused when the screen disagrees. The better mental model is a two-market basis trade with a conversion option that only an authorized participant can exercise, in a creation unit size that is big enough to matter. If the premium or discount does not clear the all-in friction, nobody has to close it, and the tape can sit there looking wrong.

I’ve also seen the most expensive mistakes come from bad clocks, not bad math. Invesco’s time-zone mismatch example is exactly the kind of thing that creates a fake nav premium that looks like free money. When the underlying market is closed or illiquid and iNAV is updating every 15 seconds off proxies, the “gap” can be optical. The tradeable question is always whether the basket can be priced, sourced, and hedged right now.