The first 55 issues asked what to buy and at what price. This one asks about something usually treated as irrelevant: what happens between the moment you hit confirm and the moment you get a fill. The plumbing is not neutral — it determines whether liquidity survives a stress day, what your ETF is actually worth, and who pays your market impact.
The Principle
The bid-ask spread is not friction. It is rent you pay for filling right now. Liquidity is a state of the market, not a property of the asset.
Source · Key Passage
"The ask-bid spread is the markup that is paid for predictable immediacy of exchange in organized markets."
— Harold Demsetz, "The Cost of Transacting," The Quarterly Journal of Economics, Vol. 82, No. 1 (1968)
Interpretation
Demsetz reframed the spread from a market defect into the price of a service. It is built from order-processing costs, inventory risk, and — most importantly — adverse selection (Glosten & Milgrom, 1985): a market maker necessarily loses to informed traders and can only earn it back from uninformed ones.
The corollary is counterintuitive: spreads widening in a crisis is not greed, it is correct pricing. The more damaging institutional fact is that market makers carry no obligation to keep quoting at all.
Case Study
May 6, 2010. Per the joint SEC–CFTC report (September 30, 2010), a sell program unloaded 75,000 E-mini S&P futures contracts (roughly $4.1 billion notional) in about 20 minutes; the algorithm throttled only to 9% of contemporaneous volume, with no price limit and no time constraint. The Dow fell roughly 998.5 points intraday (about 9%) and recovered most of it within about 36 minutes.
More than 20,000 trades printed at prices over 60% away from pre-crash levels; some stocks traded at $0.01. Nothing fundamental changed in those 36 minutes. What vanished was not value. It was the quote.
Limits & Decision Checklist
The reverse also holds: for a long-term investor with very low turnover, the spread is the smallest cost line of all — far below taxes (Day 46) and far below valuation errors. Over-optimizing execution is attention spent past the decimal point. The spread only truly hurts you in one scenario: being forced to trade when liquidity has gone. And that is decided by your leverage and cash planning, not by the market.
- What is this name's average daily turnover, and what share is my order? Above 10%, work it or skip it.
- Market order or limit order? A market order hands price-setting power to the other side.
- Am I trading in the first or last minutes of the session — the widest-spread windows?
- If I were barred from selling for three months, would I still size the position this way?
Essence · Reflection
Liquidity is most abundant when you do not need it and scarcest the moment you do.
Recall the last time you were in the state of "I have to sell today." What put you there — leverage, redemptions, or emotion?
The Principle
There is no legal guarantee tying an ETF's share price to the value of its holdings. The only thing pinning them together is the authorized participant's incentive to arbitrage. Raise the cost of that arbitrage and the pin loosens.
Source · Key Passage
"The key criterion isn't 'can you sell it?' It's 'can you sell it at a price equal or close to the last price?'"
— Howard Marks, Oaktree memo Liquidity (March 2015)
Interpretation
An ETF has two markets. The secondary market is you and other investors swapping shares that already exist — the underlying holdings are never touched. The primary market is open only to authorized participants (APs), who deliver a basket of securities in exchange for newly created shares, or redeem in reverse.
All the price discipline comes from the latter, and it requires three conditions at once: the underlying is tradable, the AP has capital and borrow, and the creation/redemption channel is open. When the underlying has stopped trading, the ETF price is not "wrong" — it is the only price still updating.
Premium (price > NAV)The AP buys the underlying basket, exchanges it for new shares, and sells them. Supply rises; the premium closes.
Discount (price < NAV)The AP buys discounted shares, redeems them for the underlying, and sells that. Shares are destroyed; the discount closes.
What it requiresThe underlying must trade, price, and borrow; the AP needs capital and risk limits; the channel must stay open. Lose one and it fails.
What failure looks likePremiums or discounts that widen and do not close. You are no longer trading NAV — you are trading whatever a willing buyer will pay.
Arbitrage is a business, not a law of physics: it stops when it stops paying.
Case Study
On March 12, 2020, the largest investment-grade corporate bond ETF (LQD) closed at a discount of roughly 5% to its NAV, with comparable discounts across credit ETFs. The fault was not the ETF: the underlying corporate bonds had almost no live quotes, and NAV was struck off stale or model prices. On March 23 the Federal Reserve announced the Secondary Market Corporate Credit Facility and said explicitly that it would buy investment-grade corporate bond ETFs. Discounts closed fast.
The argument that followed is worth remembering: was the ETF broken, or did it discover the real price ahead of NAV? Most research favors the latter — the discount measured what it would actually cost to sell those bonds today; NAV measured the pretense that yesterday's price still held.
Limits & Decision Checklist
The mechanism is weakest exactly where liquidity mismatch is worst: an intraday-tradable share wrapped around emerging-market equities, municipal bonds, or bank loans. Leveraged and inverse products are a separate hazard — on February 5, 2018, VIX spiked in a single session and the XIV short-volatility note lost over 90% of its indicative value in a day; the issuer called it for early redemption under the terms of the prospectus. A product can terminate before you get the chance to sell. That is documentation risk, not market risk.
- Can the underlying still trade on a stress day? If not, a discount is inevitable, not surprising.
- Am I looking at price or NAV? What is this fund's historical premium/discount range?
- Physical replication or synthetic (swap-based)? If synthetic, who is the counterparty?
- Do the terms allow early termination, mandatory redemption, or leverage reset?
Essence · Reflection
A wrapper does not create liquidity. It only redistributes who bears the cost of its absence, and when.
Open the holdings page of every ETF you own and ask: if everyone redeemed tomorrow, what exactly would the AP have to sell?
The Principle
HFT did two opposite things at once: it drove the explicit costs of small trades to historic lows, and it turned large orders into signals that can be front-run. Which side you are on decides whether it is net good or net bad for you.
Source · Key Passage
"The United States stock market, the most iconic market in global capitalism, is rigged."
— Michael Lewis, Flash Boys (2014)
Interpretation
That sentence is the book's most forceful and the one most in need of unpacking. Two opposite strategies have to be separated: market-making (continuous two-sided quotes, earning a razor-thin spread on turnover) and predatory (latency arbitrage, detecting and front-running large orders). The first compresses costs; the second transfers them.
The data sides with institutional evolution: the US minimum tick fell from 1/8 of a dollar (12.5 cents) in 1997 to 1/16, then to a penny with decimalization in 2001, and retail commissions went to zero. Low-turnover long-term investors are the largest net beneficiaries. What was taken came from large institutional orders that must be worked in pieces — they pay market impact, and market impact appears on no fee schedule anywhere.
Case Study
Virtu Financial disclosed in its March 2014 IPO filing that it had exactly one losing trading day out of 1,238 trading days from 2009 through 2013. That is not forecasting ability. It is a minuscule per-trade edge multiplied by an enormous number of trades — structurally the same mathematics as Medallion (Day 36), several orders of magnitude shorter in time scale.
The institutional answer came from the other side: IEX used a 350-microsecond speed bump (a coil of fiber) to neutralize latency advantage, and the SEC approved it as a national securities exchange in June 2016. Regulators thereby conceded that a speed differential can be extraction rather than service.
Limits & Decision Checklist
Treating HFT as the root of all evil is equally untenable: spreads and explicit costs fell steadily through its expansion, and there is no evidence that total trading costs rose for long-term investors. What it genuinely did was make extreme moments thinner — quotes can be pulled in milliseconds (back to Card 1). The conclusion for an individual is almost boring: you cannot compete on speed, and you do not need to. Your edge is the time horizon; HFT holds for seconds.
- What was my turnover this year? The lower it is, the less microstructure matters to me.
- Who sets my price — me or the counterparty? The limit order is the only answer.
- The "slippage" I complain about: how much of it is my own order size?
- Have I traded more often in pursuit of "better execution"? That trade is net negative.
Essence · Reflection
On a battlefield measured in milliseconds you lose by default. Fortunately, it was never your battlefield.
Multiply last year's number of trades by your average trade size, estimate what you paid for immediacy, and ask whether it was worth it.
The Principle
Index reconstitution manufactures a class of buyers and sellers who do not look at price. Their existence proves something the textbook says should not exist: demand curves for stocks slope down.
Source · Key Passage
"The evidence supports the hypothesis that demand curves for stocks slope down."
— Andrei Shleifer, "Do Demand Curves for Stocks Slope Down?", The Journal of Finance, Vol. 41, No. 3 (1986)
Interpretation
Standard theory says stocks have countless close substitutes, so demand curves should be nearly flat — buying a bit more should not move the price. Shleifer built a clean natural experiment out of S&P 500 additions: inclusion carries no new fundamental information, and the only change is that passive money must buy. He measured an average abnormal return of about 2.79% on the announcement day.
The mechanism lies in the passive manager's objective function: tracking error is the only metric they are graded on, which makes them insensitive to price and extremely sensitive to timing. The counterparty collects precisely the premium for "this must print by today's close" — the same rent as the spread in Card 1.
Case Study
Tesla's addition to the S&P 500 was announced on November 16, 2020 and took effect on December 21. The stock rose roughly 70% between announcement and inclusion; in the closing auction of December 18, over 200 million Tesla shares changed hands — the largest single closing cross in US market history at the time. Index funds had to buy at that price. They do not have the right to say it is too expensive.
The heavier lesson is twenty years older: JDS Uniphase was added to the S&P 500 in July 2000, near the top of the internet bubble, and subsequently fell more than 99% — indexation distributed a bubble valuation, mechanically and indiscriminately, to every passive holder.
Limits & Decision Checklist
But the effect is disappearing. Greenwood and Sammon's work (The Disappearing Index Effect) documents announcement-day abnormal returns for S&P 500 additions decaying from significantly positive in the 1980s and 1990s to roughly zero from the 2010s onward — arbitrageurs position ahead of it and index construction has grown more transparent. It is a clean specimen of a strategy killed by its own fame. So the takeaway is not "you can front-run the index," but this: part of the price you paid inside your passive holdings was set by mechanics, not by value.
- What are the reconstitution rules and frequency of the index my fund tracks? Have I read them?
- Is the day I plan to trade a quarterly rebalance day or an options expiry?
- Does the move I am looking at have a purely mechanical explanation?
- Is this fund now large relative to the daily volume of its underlying constituents?
Essence · Reflection
The cost of passive investing is not written on the fee schedule. It is written into the closing price on rebalance day.
Look up the date and turnover of your main index fund's last reconstitution. What did you pay to keep tracking error at zero?
Going Deeper
If an ETF is only a wrapper, why do people call it a systemic risk?
The worry is not the wrapper but the combination of scale and mismatch. When an intraday-tradable product holds a market that genuinely trades in weekly rhythm — high yield, bank loans — it promises something the underlying cannot deliver. The stress-day path is mechanical: shares are dumped, the discount widens, APs redeem, the underlying is force-sold, more selling follows. In March 2020 that chain ran halfway before a central bank balance sheet interrupted it. One discipline is enough: do not buy a product whose underlying cannot trade on a stress day with money you might need on a stress day.
How much microstructure does an individual investor actually need?
Three things. First, always use limit orders. Second, avoid the first and last ten minutes of the session, where spreads are widest and quotes least reliable. Third, lower your turnover, which cuts spread, impact, and tax at the same time (Day 46). Past that, marginal value decays to zero fast: the returns of microstructure belong to those trading a thousand times a day, while its risk belongs to everyone. Defend against the second; do not chase the first.
Where do algorithms and AI take execution next?
Most likely more of the same direction: narrower spreads and thinner depth in normal times, higher correlation in extreme ones. As more quotes are generated by similar models, they withdraw under similar conditions — falling diversity is the classic source of fragility in complex systems (Day 25). The implication for a long-term investor gets clearer, not murkier: as everyone's time horizon shortens, capital willing to hold for years becomes scarcer, and scarcity is where premia come from.
If passive money ignores price, who sets prices in the end?
The marginal trader — and passive money almost never trades at the margin, since its turnover is minimal and its purchases simply settle in. Pricing power still sits with active capital; that hand is just getting smaller (Day 44). The real risk is not "nobody prices it" but that below some threshold of price-setters, the reaction to new information turns sluggish first and abrupt afterward. That points to the same conclusion as the first three cards: an appearance of stability is often held up by forces that can withdraw instantly.