ETFs, liquidity profile and valuation issues

ETFs are generally seen as being a low risk way to invest in an asset class.  This is often stated to be because the assets within the ETF are valued by the market and the ETF manager

[I recommend reading an article on this topic in the Economist of 25^th October 2014]

 

I broadly agree with the assertion that the Economist makes in conclusion:

“Even in the case of corporate-bond funds, it is hard to see how things could go badly wrong. The total assets of corporate-bond ETFS are $141 billion, compared with annual bond issuance of £3.2 trillion. In other words, ETFs comprise only a tiny part of demand for the asset.  Unlike banks or hedge funds, most ETFs do not use borrowed money, or leverage.  There will doubtless be individual ETFs that get into trouble in the future; there will probably be scandals.  Some will suffer spectacular falls in value, just as technology funds plunged in the early 2000s. But that does not make them a systemic threat.”

But I believe with further research it would be possible to minimise a certain type of risk to a holder of an ETF.  Let’s go into more detail..

Firstly, let’s create a taxonomy of ETF strategies.  If we work on the basis that an ETF is tracking an index which is replicable then there are a number of ways to do this:

1. Full index replication.  In this case, an ETF holds the individual constituents of the index in as close as possible resemblance to the index weights.  So if equity X is 2.70045% of index then this style of ETF will attempt to hold that a weight of 2.70045% in the ETF. Note that it’s pretty much impossible to actually dynamically maintain true index replication, since often cash inflows and outflows mean that absolute replication would result in lots of odd-lot trading as positions are built-upon or top-sliced.  So this is an aspiration rather than a true state of the world.


2. Stratified sampling.  Rather than hold every constituent, get some quantitative analysts to look at the variance/co-variance matrix for the index.  Typically you can see that some constituents move more or less in line with the index and so by careful portfolio construction it should be possible to satisfy for two constraints – to ensure that the ETF valuation moves in line with the index and to minimise trading costs.


3. Derivatives based replication.  In this model, the ETF enters into a series of derivative contracts with a broker.  Typically the broker is part of the same group that manages the ETF.  This type of ETF has been under a lot of scrutiny, since there is a lot of debate about whether this is actually a low-risk investment at all – what happens if the broker that is the other side to the derivative contracts goes bust? We are not going to look at this type of ETF in the rest of this post.

For the following analysis, I do not mention any accrued management fees, but they can be safely ignored since they neither add nor detract from explanatory power of this model.

Let’s look at a hypothetical corporate bond ETF that is started with a lump sum investment.  For clear reasons this does not resemble any real ETF of which I am aware. Our hypothetical ETF will be called “Sampled Replication” and attempts to match a particular index in terms of sector exposure, duration, income profile and so on.  Within that index replication strategy the hypothetical ETF provider decides that there is no need to have full index replication and so some Fixed Income quants put together a model in Matlab, SAS, S+, R, Python or whatever the cool kids are using and work out a replication model.

The model output is given to the traders and they start to build the portfolio.  At some point, victory is declared and the ETF is ready to go.

Now, let’s look at the modelling that the quants used.  Again, this is just hypothetical, but it is not outlandish.  The quants were asked to look at the following:

• Concentration risk – don’t invest more than x% of the ETF in one bond.
• Concentration risk – don’t invest more than y% of the ETF in one bond issuer.
• Concentration risk – don’t invest more into one bond an amount more than z% of the outstanding value of the bond.
• Duration match – match the duration of the index
• Yield match – match the yield of the index
• Cash minimisation – keep cash to the minimum possible over the index weight in cash.  If the index has a zero cash weight, aim for zero cash. Do not go short cash – since most ETFs cannot go overdrawn since that is leverage.
• Quality match – on a weighted average rating basis, match the index
• Within the above constraints – minimise the number of positions in the portfolio  

So, the outcome will be a portfolio of bonds that in static analysis will behave rather like the underlying index.

Now, let’s introduce a new metric – liquidity profile for the overall ETF portfolio.  If we look at equity markets a quantitative analyst or trader will be familiar with the concept of average daily volume (ADV).  When looking to sell a position in an equity a trader will look at the size of the proposed trade as a percentage of the average daily volume.  If the %ADV is small then the position can be sold without causing as much market impact (reduction in execution price at execution as opposed to the last traded price before execution commenced – this is usually referred to as implementation shortfall and there is a swathe of literature on that topic to which I see no value in adding) as a large %ADV order.  The concept of ADV requires timely, consolidated information from every execution venue.  This is something that every equity trading desk needs in order to trade blocks, programs or algorithms successfully.

Within the Fixed Income world the concept of ADV is not meaningful, since there are so few trades that a measure of volume on a daily basis is only useful for the most liquid assets such as US T-bonds, T-bills and T-notes. For nearly all corporate bonds ADV is not a powerful concept to explain trading behaviour.

And at this point we hit the problem….

Less positions for a given size of fund must mean that the quantitative model based stratified sample fund has larger positions in each individual bond versus a true index replication fund.

So, let’s look at the two different ETFs:

“Full Replication”, hereafter “FR”

Full index replication of index with N constituents – hold N bonds in very close proximity to the index weights along with some always positive cash constituent.  ETF is valued at $z billion. Average position is of course (z-cash component)/N. At inception W shares are issued.  Therefore the price per share is $z/W

“Sampled Replication”, hereafter “SR”

Stratified sample ETF of index with N constituents – holds N-x bonds with only limited resemblance to index weights along with some always positive cash constituent.  ETF is valued at $z billion. Average position is of course (z-cash component)/N-x. At inception W shares are issued.  Therefore the price per share is $z/W

Clearly the larger x, the larger the average size of position.

Now, let’s apply a market shock to the system.  There is a financial crisis of some kind and investors rush to quality and cash, selling out of positions in everything else.  That’s not far from what happened during the dark days around the collapse of Lehman Brothers.  So – investors in FR and SR both head for the exits.

Investors place orders with their brokers and the brokers look to internalise this flow.  Let’s imagine that in this scenario the market is pretty one sided – lots of offers and few bids.  As such orders end up with the ETF manager as calls for redemption.  Now the ETF manager has to redeem units.  There are a number of ways to do this, depending on the way the ETF was set up.

1. The ETF manager sells assets to raise cash and delivers cash to the broker.
2. The ETF manager offers a total ‘in-specie’ redemption.  In this case the broker receives a slice of each holding in the portfolio.
3. The ETF manager offers a sampled ‘in-specie’ redemption.  In this case the broker receives a slice of many holdings in the portfolio.

Now, let’s simplify this.  Whether the authorised participant is given a sampled or total in-specie redemption in place of a cash redemption is simply moving work around from the ETF manager to the authorised participant.  So, there has to be a sale by someone of some of the ETF underlying assets to raise cash. For either “Full Replication” or “Sampled Replication” the three choices are all valid.

But – as we already know, “SR” has a large average position size versus “FR”.

Now – let’s overlay that statement with what we know about corporate bond trading in the period before the collapse of Lehman Brothers and afterwards…

Before the collapse of Lehman Brothers banks would allocate a chunk of capital to corporate bond trading.  This capital would be used to fund buying positions from client firms and running that position until it could be sold.  In some cases positions would be held for a long time.  The model was very much around using the balance sheet of the bank to take positions off buy-sides and then sell that position on, either directly in the secondary market or through an interdealer broker.

After Lehman Brothers and the swathes of regulatory edicts that have come to fruition we have seen that liquidity in corporate bonds has diminished as banks have pulled capital from unprofitable flow trading businesses to either bolster regulatory capital or other allocate to other business units.  So, there is not a lot of liquidity in corporate bonds.

When a firm tries to sell a position into an illiquid market – what happens?  Simple supply and demand suggests that an increase in supply requires a reduction in market price until a market clearing price is found and an equilibrium is restored.

So in this paradigm we should see that a shareholder selling shares in “SR” will get a lower price per share than a shareholder selling shares in “FR”. Remember that the share price at inception was $z/W.  Therefore we should see “SR” shares redeeming for $z-D/W and “FR” shares redeeming for $z-d/W where d<D.

This is a naïve analysis and not one I commend to the reader.

Why?

In that naïve model, there is some level of liquidity per bond, an amount of a bond that the market is willing to buy or sell at a particular price level.  While that’s appealing for the purposes of modelling a portfolio it’s not realistic.

Why?

The banks still have capital with which they trade corporate bonds, albeit much diminished.  But the allocation of that capital is not at a per-bond level.  No bank of which I am aware would manage risk by suggesting that exposure should be managed at a bond-by-bond level. As such, the big question becomes “if liquidity in corporate bond markets is not allocated at the “per name” level then at what higher level of abstraction?”

This question is important in establishing the expected reaction of redemption price “SR” and “FR” to a market crisis in line with the events of Lehman Brothers collapse.

For “SR” and “FR” – if they each need to sell say, 25% of their portfolio with which portfolio will have a the more severe impact be seen?

If liquidity exists at the “per name” level that clearly “FR” will find it easier to sell bonds since it holds more of them with smaller positions on average than “SR”.  But if liquidity exists at the total market level then it makes no difference.  If liquidity exists at some level of abstraction somewhere between these two bounds then it would require analysis of the two ETFs at that level of abstraction.

At this point I will interject with my own experience – that at firms with which I am familiar with aspects of their trading strategies and models – their liquidity has been allocated firstly to a specific business line and subsequently to a sector and/or currency based level.  So a bank may allocate capital to their corporate bond business and within that capital pool allocate to particular currencies and sectors, so US Dollar Technology Media and Telecoms may be allocated a lot of capital and Euro Industrials may be allocated very little.

If that model holds for a broad swathe of the market then we see that the “SR” ETF may or may not have the same portfolio liquidity profile as the “FR” ETF and so the impact of trying to sell 25% of the portfolio is not knowable in advance.

So – does this post have a conclusive recommendation? No.  The only way to really figure out which ETF model can withstand a price shock is for a thorough analysis of the index replication strategy and model for a specific ETF compared to the peer group ETFs.  The look at the authorised participants and try and understand the way they allocate capital for trading.  The liquidity profile mismatch risk between the ETF and the market can then be measured.  However, good luck with trying to get a bank to confess how they allocate capital to corporate bond trading...

Furthermore, bear in mind that in the event of severe market disruption, all bets are off as a flight to quality can trump any considerations around provision of an orderly market.

In summary, as for so many things in the world of finance, “caveat emptor" and remember that not all ETFs are created equal...