2. Inside an ETF: physical and synthetic replication

Costs 4: Funds, ETFs and products

An ETF trades on a market, so its price is set by the supply and demand of that market. And yet a large ETF on a liquid index always trades at a price practically identical to the value of the portfolio inside it. That is neither a coincidence nor a promise from the fund manager: it is the product of a specific mechanism worth understanding, because it is when that mechanism jams that the surprises turn up.

The problem shows up clearly in a closed-end fund, that investment company with a fixed number of shares which can trade for years at a 15% discount to its assets with nothing to correct it: there is no way of turning its share into the underlying holdings, so there is no way of forcing convergence. The ETF made that exchange possible.

Creation and redemption of units

Alongside the fund manager work firms known as authorized participants: investment banks and market makers holding a contract that lets them do something you cannot.

• Creation. The authorized participant hands the fund manager the basket of securities that makes up the index, in exactly the right proportions, and receives in exchange a large block of newly issued ETF shares (usually tens of thousands of shares to a block).
• Redemption. The reverse process: it hands back a block of shares and receives the basket of securities.

With that, any divergence between the ETF's price and the value of the portfolio turns into an arbitrage. If it trades dear, the participant buys the index's shares, delivers them, receives ETF shares and sells them, pocketing the difference and pushing the price down. If it trades cheap, it does the opposite. The price does not stay pinned to net asset value out of goodwill, but because closing the gap pays somebody, right up until it stops paying.

The corollary matters: the anchor is only dependable if the underlying holdings can be bought and sold. With large shares it is close to perfect; with illiquid bonds, or an emerging market closed because of the time difference, the arbitrage is approximate, and visible premiums and discounts appear on precisely the days of stress.

Physical replication: full and by sampling

The direct way to track an index is to buy its constituents. It is done in two ways:

• Full physical replication. The fund holds every security in the index at its exact weight. This is the norm for indices with few constituents and plenty of liquidity.
• Optimized sampling. The fund buys a subset chosen so that the portfolio behaves statistically like the index, leaving out the tiny or illiquid names in the tail. A world index of more than 3,000 constituents is almost always replicated this way.

Sampling is not a swindle, it is a decision about cost. But the portfolio and the index stop being identical, and when small companies behave differently from the rest, the difference shows.

Synthetic replication: the swap

The alternative is not to buy the index at all. The fund holds a collateral portfolio (which may have nothing whatever to do with the index) and signs a swap with a bank, a *total return swap* to give it its full name: the fund hands over the return of that portfolio and the bank undertakes to deliver the total return of the index, dividends included, in exchange for a fee.

The advantages are real: tracking is usually tighter, because the bank absorbs the deviation, and for indices that are awkward to reach it can be the only clean way of doing it. There is a tax advantage too, which the next module develops.

Counterparty risk and how it is contained

Here is the obvious catch: if the bank goes under, the promise is worth whatever its insolvency estate is worth. That is counterparty risk, and European regulation of harmonized funds caps it at 10% of net assets against any single counterparty. It is contained in practice in three further ways: a portfolio of collateral is pledged in the fund's favor, normally above 100% of the exposed value; valuation haircuts are applied to that collateral according to its quality; and the swap is reset frequently, settling the accumulated difference so that the exposure spends almost all its life close to zero. Many funds use several counterparties as well.

Even so, two risks do not go away: that the collateral is of poorer quality than its label suggests, and that the bank fails precisely when the market is falling, which is when the collateral is worth less too.

Securities lending inside the fund

A physical ETF may lend its shares to third parties, typically to short sellers, in exchange for a fee and against collateral. That is a different matter from your broker lending yours: here the lender is the fund, and the income belongs to the fund's assets, which is to say to the unitholders.

In practice it is shared out. Some fund managers return all or nearly all of it to the fund; others keep a declared portion, and a split of around two thirds to the fund is common among the large issuers. The figure appears in the prospectus and is worth a look: that income can offset a good part of the TER. The added risk is real though small: if the borrower fails to return the securities, the fund keeps the collateral and depends on it covering the loss.

A real case: the scare of 2011

In 2010 synthetic replication came to account for close to half of European ETFs by assets. In April 2011 the Financial Stability Board published a note pointing out that the tangle of swaps between funds and banks within the same group could transmit strain into the system, above all because the collateral posted did not always bear any relation to the index being replicated.

Nobody went under and the mechanism held, but the effect on the market was considerable: several fund managers converted synthetic products into physical ones and began publishing the daily composition of the collateral. The synthetic share in Europe fell from that region of 45% to a distinctly smaller fraction. The lesson was not that synthetic replication is unacceptable, but that a product can be cheap in TER and expensive in a risk the TER does not capture.