How ETFs Work: Creation, Redemption, and Arbitrage

Direct Answer

An ETF's market price stays close to NAV because authorized participants (APs), large broker-dealers with contractual rights to deal directly with the fund, continuously arbitrage away divergences. When the ETF trades at a premium to NAV, APs buy the cheaper underlying basket, deliver it to create new ETF shares, and sell those shares on-exchange at the premium. When the ETF trades at a discount, APs buy discounted ETF shares and redeem them for the more valuable underlying basket. This loop runs continuously during market hours and typically keeps the premium or discount for liquid equity ETFs within 1-5 basis points. The mechanism can break down when underlying securities are illiquid, restricted, or in a different time zone, or when APs reduce their arbitrage activity during market stress.

This mechanism is specific to ETFs. For how it compares with mutual fund share pricing at NAV, and the broader fund category, mutual funds, index funds, target-date funds, and money-market funds, see Swoopr's Mutual Funds & Index Funds hub.

Key Takeaways

• ETF shares trade on exchanges between buyers and sellers; authorized participants create and redeem shares in large blocks (creation units) directly with the fund.
• The arbitrage mechanism, not regulation, is what keeps ETF prices near NAV. It works because APs profit from exploiting any gap between ETF price and NAV.
• In-kind creation delivers a basket of securities to the fund; in-kind redemption receives a basket from the fund. Cash creation/redemption is used when in-kind transfer is impractical.
• Most retail investors never interact with the creation/redemption mechanism. They buy and sell ETF shares on the secondary market, priced by supply and demand.
• The creation basket disclosed by the ETF sponsor each morning tells APs exactly which securities to assemble; some ETFs use a "substitute basket" if the actual portfolio can't be disclosed (active non-transparent ETFs).
• Premiums and discounts can widen and persist when underlying liquidity evaporates, when markets are closed (international ETFs), or when creation/redemption costs rise above the spread being exploited.

Core Concepts

The Primary Market: Creation and Redemption

An ETF has two simultaneous markets. The secondary market is the exchange where retail and institutional investors buy and sell existing ETF shares throughout the day at prices set by supply and demand. The primary market is the direct channel between authorized participants and the ETF sponsor, where new shares are created and existing shares are retired.

In the primary market, creation and redemption happen in blocks called creation units, typically 25,000 to 200,000 shares. An AP wanting to create a creation unit assembles a basket of securities matching the ETF's required composition, delivers it to the fund's custodian, and receives an equivalent block of ETF shares in return. The AP then sells those shares on the exchange. Redemption works in reverse: the AP buys a creation unit's worth of ETF shares on the exchange, delivers them to the fund, and receives the corresponding basket of securities. Both transactions are settled through the standard securities clearing system.

The composition of the creation basket is published each morning by the ETF sponsor or an index provider. For index-tracking ETFs, this basket closely mirrors the fund's portfolio. For active non-transparent ETFs (NTFs), a "substitute basket" of highly liquid proxy securities stands in for the actual holdings, protecting the manager's trade information while still enabling arbitrage.

How Arbitrage Keeps Prices in Line

The mechanism works because of a simple profit opportunity. If an ETF's market price rises above the value of its underlying holdings (a premium), an AP can buy the cheaper underlying basket, deliver it to create ETF shares worth $100 NAV, and immediately sell those shares at the $100.05 premium, capturing $0.05 per share, minus transaction costs, before anyone else can react. The act of selling ETF shares pushes the price back toward NAV; the buying of underlying securities pushes their prices up toward the ETF's implied price. The arbitrage closes itself.

At a discount, the mirror trade: the AP buys cheap ETF shares on the exchange (at $99.95, say), redeems them for a basket worth $100 NAV, and sells the individual securities, capturing the $0.05 spread. The buying of ETF shares lifts the market price; the selling of underlying securities moderates their prices. Again, the divergence collapses.

In practice, this arbitrage is so fast and competitive that for liquid equity ETFs tracking well-known indexes, premiums and discounts rarely exceed 5 basis points (0.05%) during regular market hours. The spread is essentially the transaction cost floor, the minimum the gap needs to be for arbitrage to be economically worthwhile.

The Role of Market Makers

Authorized participants often also act as market makers in the ETF's secondary market. They provide two-sided quotes, bid and ask prices at which they are willing to buy and sell ETF shares, and profit from the bid-ask spread. As market makers, they continuously update their ETF quotes based on the current value of the underlying portfolio, their inventory position, and their hedging costs.

When a large institutional order hits the market, a market maker may fill it from inventory rather than immediately creating or redeeming shares. They manage this inventory risk by hedging against the ETF's underlying, often using futures, options, or a correlated basket. The creation/redemption mechanism is the backstop they use to square their books at end of day or when inventory grows too large. This two-tier structure, intraday market-making supplemented by end-of-day creation/redemption, is why ETF trading is efficient even in high-volume conditions.

Cash vs. In-Kind Creation and Redemption

Most equity ETFs use in-kind creation and redemption, where physical securities change hands between the AP and the fund. This is the structure that gives ETFs their famous tax efficiency: when low-basis securities are redeemed in-kind, the fund does not sell them and therefore does not realize a taxable gain. The embedded gain transfers to the AP, which bears it when it eventually disposes of those securities. The fund's remaining shareholders are shielded.

Some ETFs must use cash creation and redemption, for example, ETFs holding commodity futures (which can't be in-kinded), ETFs in markets where in-kind transfer is legally restricted, or cryptocurrency ETFs (where custodial and regulatory constraints have historically required cash creation). Cash-create ETFs are less tax-efficient because the fund must sell holdings to raise cash for redemptions, potentially generating realized gains that flow to all shareholders. They are functionally closer to mutual funds in their tax behavior.

Worked Scenario

1. Morning setup: An ETF tracking a 500-stock index opens with a NAV of $50.00 per share, calculated using last night's closing prices. The iNAV (intraday NAV) ticks every 15 seconds based on live prices of the underlying stocks.
2. Premium develops: After a positive economic report at 10:00 AM, buyers flood into the ETF. Heavy buying pressure pushes the ETF's market price to $50.08 while the underlying stocks have only risen to an implied NAV of $50.03. A 10-basis-point premium exists.
3. AP spots the opportunity: Market maker XYZ, acting as AP, calculates the arbitrage: buy $50.03 worth of the creation basket securities in the open market, deliver them to create ETF shares with $50.03 NAV, sell those shares at $50.08. Gross profit per share: $0.05. Transaction costs (commissions, market impact, settlement): $0.01. Net profit: $0.04 per share.
4. Creation units in play: The creation unit size is 50,000 shares. XYZ creates 2 creation units (100,000 shares), netting approximately $4,000 in arbitrage profit.
5. Price convergence: XYZ's selling of 100,000 ETF shares at $50.08 increases supply, pushing the ETF price down. Its buying of underlying securities pushed those prices up slightly. Within minutes, the premium has narrowed to 1-2 basis points.
6. Discount scenario: In a market selloff at 2:00 PM, panic sellers hit the ETF harder than the underlying. ETF falls to $49.80 while NAV shows $49.87. APs now reverse the trade: buy cheap ETF shares at $49.80, redeem for basket worth $49.87, sell the individual stocks. The discount disappears within minutes.

Measurement Framework

Common Failure Modes

International Time-Zone Mismatch

An ETF holding Japanese equities trades on U.S. exchanges while the Tokyo Stock Exchange is closed. The ETF's NAV is calculated using yesterday's closing prices for the Japanese holdings, but the ETF's market price reflects real-time information about events affecting Japan (yen moves, overnight futures, macro news). Arbitrage cannot work because APs can't buy the underlying Japanese stocks during U.S. hours, the Tokyo exchange is closed. The result is that the ETF price can diverge significantly from its official NAV, while actually more accurately reflecting the current fair value of its holdings. Investors who see a "5% premium" on an international ETF after major overnight news are typically seeing the market price correctly adjust before the NAV catches up at the next Tokyo open.

Illiquid Underlying Markets

High-yield bond ETFs, municipal bond ETFs, and other fixed-income vehicles face a structural challenge: the underlying bonds are often thinly traded or dealer-only markets, while the ETF shares trade continuously on an exchange. During normal conditions, market makers can hedge their ETF positions through correlated bond futures or baskets. During stress, when bond markets freeze and bid-ask spreads widen to multiple points, the cost of assembling the creation basket exceeds the value of any arbitrage profit. APs step back, and the ETF can trade at discounts of 2-5% to NAV, as was observed in March 2020 for some high-yield and investment-grade bond ETFs. Paradoxically, the ETF price in these cases often reflects the true distressed market value of the bonds more accurately than the stale NAV that uses dealer-quoted (not executed) prices.

Regulatory or Settlement Constraints

Some underlying markets have rules that impede in-kind transfers. ETFs holding securities from markets with restrictions on foreign ownership, markets using different settlement cycles, or assets with special custody requirements may use a cash creation/redemption structure or a custom in-kind basket that excludes the problematic securities. When this basket diverges from the actual portfolio, arbitrage is imprecise, and premiums/discounts may be structurally wider than for simpler funds.

Authorized Participant Risk

ETFs rely on APs to participate in arbitrage voluntarily. APs have no obligation to create or redeem shares, they do so when it's profitable. If APs face capital constraints, counterparty risk concerns, or simply choose to reduce risk during market turmoil, they may scale back or pause arbitrage activity entirely. This is rare for major equity ETFs with multiple competing APs, but for niche or small-AUM ETFs with only one or two APs, this risk is more concentrated. Investors should check how many authorized participants an ETF has; funds with one or two APs are more exposed to AP risk than those with 10 or 20.

Active Non-Transparent ETF Substitution Basket Drift

Active non-transparent ETFs use a substitute basket, a portfolio of proxy securities designed to correlate with the actual portfolio without revealing active positions. If market conditions cause the substitute basket to diverge from the actual portfolio's performance, AP hedging becomes imprecise, and the arbitrage mechanism can generate premiums or discounts even under normal conditions. Some active NTF structures include periodic "proxy overlap" tests to ensure the substitute basket remains sufficiently correlated, but these constraints are not perfect.

Why an ETF's Own Volume Is Not Its Liquidity

The creation and redemption process described above has a consequence that is routinely missed: an ETF's tradable size is not set by how many of its shares change hands. It is set by how easily an authorized participant can assemble or unwind the basket underneath. A fund that trades 20,000 shares a day and holds mega-cap equities can absorb a far larger order than a fund that trades 500,000 shares a day and holds thinly traded credit, because the AP behind the first fund can hedge and source the basket instantly while the AP behind the second cannot.

This is the single most practical reason to understand the primary market. Secondary-market volume tells you about the queue of existing buyers and sellers. Primary-market capacity tells you how much new supply can be manufactured on demand, and that is what determines whether a large order moves the price.

The Two Layers of ETF Liquidity

Every ETF has liquidity at two levels, and they can be very different from each other.

For a large order the second layer usually dominates. When an investor wants more shares than the order book can supply at a reasonable price, a market maker does not simply refuse. It quotes a price at which it can source the basket, hedge the exposure, and deliver the shares through creation. The price it quotes therefore reflects the cost of trading the underlying securities, not the depth of the ETF's own order book.

What Average Daily Volume Actually Measures

Average daily volume in ETF shares is a measure of how much trading has already been happening, which is a statement about past demand rather than about capacity. It is genuinely useful for one thing: the cost of a small order that can be filled from the existing book without engaging a market maker's inventory or the creation process. Beyond that size, it stops being the binding constraint.

Three specific misreadings follow from treating it as the whole answer:

• Rejecting a newly launched fund because it has low volume. A fund that launched last month has almost no trading history by construction. If it holds liquid large-cap equities, its tradable size on day one is a function of those equities, not of its own history.
• Assuming a high-volume fund is easy to trade in size. Volume can be high because the fund holds something volatile and heavily traded by short-horizon participants, while the underlying market it must source from is thin.
• Comparing two funds on volume alone. Volume is denominated in shares, so a fund with a low share price shows a larger share count for the same dollar value. Convert to dollar volume before any comparison, and even then treat it as the smaller half of the question.

Reading the Liquidity Signals Rule 6c-11 Already Requires

The disclosure conditions attached to Rule 6c-11 mean an ETF already publishes several liquidity signals on the same free public website as its holdings. They are more informative than volume and are consistently overlooked.

• Median bid-ask spread. The rule requires it to be expressed as a percentage rounded to the nearest hundredth and computed in a specific way: identify the fund's national best bid and national best offer as of the end of each 10 second interval during each trading day of the last 30 calendar days, divide the difference between each bid and offer by the midpoint, and take the median of those values. Because the method is prescribed, this figure is comparable between funds in a way that a quoted spread observed at one moment is not.
• Premium or discount as of the prior business day, alongside net asset value per share and market price.
• A premium and discount history table and line graph covering the most recently completed calendar year and the completed quarters since, or the life of the fund if shorter. A fund that has spent many days outside a narrow band is telling you something about how reliably its arbitrage mechanism operates.
• A required explanation for sustained dislocation. If the premium or discount is greater than 2 percent for more than seven consecutive trading days, the fund must post a statement to that effect and a discussion of the factors reasonably believed to have materially contributed to it.

That last item is the closest thing to a public alarm that the wrapper is under strain, and it is disclosed by the fund itself. How to read an ETF holdings file covers the same page's holdings disclosure, which is where the basket that drives all of this becomes visible.

Worked Example: Two Funds, Same Order, Different Outcome

Hypothetical example, for education only.

Both funds below are hypothetical and were constructed for this guide. An investor wants to buy 5,000,000 dollars of one of them.

Read on the ETF's own volume, Fund Y looks five times more liquid and Fund X looks like a fund you should not put five million dollars into. Read on the basket, the conclusion inverts: the Fund X order is a rounding error in the underlying market, while the Fund Y order asks each underlying holding to absorb more than a quarter of a typical day's trading. Every figure was computed for this illustration from the stated inputs: 5,000,000 divided by 500 holdings is 10,000 per name, 5,000,000 divided by 60 holdings is 83,333 per name, and 83,333 against 300,000 of daily volume is 27.8 percent.

The practical translation is that Fund X can very likely be bought at a spread close to the underlying large-cap spread, using a limit order and a market maker who will create shares to fill it. Fund Y will be quoted a spread that reflects the cost of accumulating 60 illiquid positions, however active its own screen looks.

When the Wrapper Is Less Liquid Than It Looks

The reverse case is the more dangerous one, because it is the case where an investor believes they hold something liquid. An ETF trades continuously and settles like a stock, which creates an impression of liquidity that the underlying market may not support. The wrapper is thinner than it appears when several of the following are true at once:

• The underlying trades by appointment. High-yield credit, bank loans, municipal bonds, frontier equities, and small-cap securities in general are dealer markets or thinly quoted markets, so the basket cost is not observable from a screen.
• The fund is a large share of the underlying float. If the fund owns a meaningful percentage of the outstanding shares of its holdings, redemption in size means selling into a market where the fund is the dominant seller.
• The AP list is short. Arbitrage is voluntary. A fund with one or two authorized participants depends on those firms choosing to participate.
• The fund uses cash creation and redemption. Cash creation transfers the trading cost and the execution risk from the AP to the fund, which means the cost lands on remaining shareholders rather than on the transacting party.
• The premium and discount history shows repeated wide days. The required table and graph are the record of how the mechanism has behaved, not a forecast, and a fund that has dislocated repeatedly in calm conditions is unlikely to hold together in stressed ones.
• The fund's own liquidity classification is unknown to you. Under Rule 22e-4 a fund classifies each portfolio investment as a highly liquid, moderately liquid, less liquid, or illiquid investment, using thresholds defined in business or calendar days to convert to cash without significantly changing the investment's market value. Those classifications are reported to the SEC on Form N-PORT, and Form N-PORT's general instructions state that the SEC does not intend to make the liquidity classification data public in a form identifiable to any particular fund. You can estimate the answer from the holdings; you cannot read the fund's own.

None of these makes a fund unusable. Together they explain why the same wrapper can behave like a stock on an ordinary Tuesday and like the underlying market on a bad one.

A Liquidity Check Before You Trade

1. Size the order in dollars, not shares, and compare it to the fund's dollar volume rather than its share volume.
2. Open the holdings file and look at what the basket contains. The liquidity of those securities is the real constraint on anything larger than a routine order.
3. Read the published median bid-ask spread, which is computed on a prescribed method and is therefore comparable across funds.
4. Read the premium and discount history, and note how often the fund has traded outside a narrow band.
5. Check whether the fund has ever had to post a sustained-dislocation explanation, which is required above 2 percent for more than seven consecutive trading days.
6. Check the creation and redemption method in the prospectus. In-kind and cash structures distribute trading costs to different people.
7. Use limit orders, and avoid the open and the close in funds whose underlying markets are not fully open or are in their own auction periods.
8. For a large order, ask about a risk price or a creation-based fill rather than working it through the displayed book, because the displayed book is the smaller half of the fund's real capacity.

The underlying-liquidity question also drives what happens to any of this in a stressed market, which is where bond ETF mechanics takes over, since fixed-income baskets are where the two layers diverge most sharply.

FAQ

References

• SEC Rule 6c-11 (ETF Rule): Final Rule, 2019
• ICI 2024 Investment Company Fact Book: ETF section
• Federal Reserve: "Can ETFs Increase Market Fragility?" Working Paper (Bhattacharya & O'Hara, 2018)
• BIS Working Paper: "Market Structures and Systemic Risks of Exchange-Traded Funds" (Ramaswamy, 2011)
• eCFR: 17 CFR 270.6c-11, Exchange-traded funds: the daily website disclosure conditions relied on above, including the median bid-ask spread computation from 10 second interval national best bid and offer observations over the last 30 calendar days, the premium and discount table and line graph, and the required explanation when a premium or discount exceeds 2 percent for more than seven consecutive trading days.
• eCFR: 17 CFR 270.22e-4, Liquidity risk management programs: the highly liquid, moderately liquid, less liquid, and illiquid investment classifications and their day-count definitions.
• SEC: Form N-PORT: General Instruction F, which states that the SEC does not intend to make liquidity classification data public in a form identifiable to any particular fund.