JOURNAL OF INVESTMENT MANAGEMENT, Vol. 7, No. 4, (2009), pp.

43–62
© JOIM 2009
JOIM
www.joim.com

THE DYNAMICS OF LEVERAGED AND INVERSE

EXCHANGE-TRADED FUNDS
Minder Cheng∗ and Ananth Madhavan∗

Leveraged and inverse Exchange-Traded Funds (ETFs) have attracted significant assets lately.
Unlike traditional ETFs, these funds have “leverage” explicitly embedded as part of their product
design. While these funds are primarily used by short-term traders, they are gaining popularity with
individual investors placing leveraged bets or hedging their portfolios. The structure of these funds,
however, creates both intended and unintended characteristics that are not seen in traditional
ETFs. This note provides a unified framework to better understand the underlying dynamics of
leveraged and inverse ETFs, their impact on market volatility and liquidity, unusual features of
their product design, and questions of investor suitability. We show that the daily re-leveraging
of these funds can exacerbate volatility towards the close. We also show that the gross return of a
leveraged or inverse ETF has an embedded path-dependent option that under certain conditions
can lead to value destruction for a buy-and-hold investor. The unsuitability of these products
for longer-term investors is reinforced by the drag on returns from high transaction costs and tax
inefficiency.1

1 Introduction end-January 2009, there are now over 106 lever-

aged and inverse ETFs in the US alone with Assets
Leveraged and inverse Exchange-Traded Funds Under Management (AUM) of about $22 billion.2
(ETFs) provide leveraged long or short exposure
to the daily return of various indexes, sectors, and
The space now comprises leveraged, inverse, and
asset classes. These funds have “leverage” explicitly
leveraged inverse ETFs offering 2× or 3× long
embedded as part of their product design. The cat-
exposure or short exposure of −1×, −2×, or −3×
egory has exploded since the first products were
the underlying index returns. The most recent prod-
introduced in 2006, especially in volatile sectors
ucts authorized by the US Securities and Exchange
such as Financials, Real Estate, and Energy. As of
Commission (SEC) offer the highest leverage fac-
tors. However, the bulk of AUM remains in 2×
∗ Barclays Global Investors, N.A. 400 Howard Street, San leveraged products. Coverage has also expanded
Francisco, CA 94105. beyond equities and includes commodities, fixed

FOURTH QUARTER 2009 43

Not for Distribution

44 MINDER CHENG AND ANANTH MADHAVAN

income and foreign exchange. There is strong activity is always in the same direction as the under-
growth in this space outside the US as well. In lying index’s daily performance. The hedging flows
addition, option contracts on leveraged ETFs have from equivalent long and short leveraged ETFs thus
also gained in popularity. Leveraged and inverse do not “offset” each other. The magnitude of the
mutual funds analogous to ETFs have also grown potential impact is proportional to the amount of
in popularity. Other than the fact that they offer assets gathered by these ETFs, the leveraged mul-
investors liquidity at only one point in the day, the tiple promised, and the underlying index’s daily
structure of these products is identical to leveraged returns. The impact is particularly significant for
and inverse ETFs and hence our analysis is fully inverse ETFs. For example, a double-inverse ETF
applicable to these funds too. promising −2× the index return requires a hedge
equal to 6× the day’s change in the fund’s Net
Several factors explain the attraction of lever- Asset Value (NAV), whereas a double-leveraged
aged and inverse ETFs. First, these funds offer ETF requires only 2× the day’s change. This daily
short-term traders and hedge funds a structured re-leveraging has profound microstructure effects,
product to express their directional views regard- exacerbating the volatility of the underlying index
ing a wide variety of equity indexes and sectors. and the securities comprising the index.
Second, as investors can obtain levered exposure
within the product, they need not to rely on While a leveraged or inverse ETF replicates a mul-
increasingly scarce outside capital or the use of tiple of the underlying index’s return on a daily
derivatives, swaps, options, futures, or trading basis, the gross return of these funds over a finite
on margin. Third, individual investors—attracted time period can be shown to have an embedded
by convenience and limited liability nature of path-dependent option on the underlying index.
these products—increasingly use them to place We show that leveraged and inverse ETFs are not
longer-term leveraged bets or to hedge their suitable for buy-and-hold investors because under
portfolios. certain circumstances the long-run returns can be
significantly below that of the appropriately lev-
The structure of these funds, however, creates both ered underlying index. This is particularly true for
intended and unintended characteristics. Indeed, volatile indexes and for inverse ETFs. The unsuit-
despite their popularity, many of the features of ability of these products for longer-term investors
these funds are not fully understood, even among is reinforced by tax inefficiency and the cumulative
professional asset managers and traders. This paper drag on returns from transaction costs related to
provides a unified framework to better under- daily re-balancing activity.
stand some key aspects of these leveraged and
inverse ETFs, including their underlying dynam- The paper proceeds as follows: Section 2 shows
ics, unusual features of their product design, their how leveraged and inverse ETF returns are related
impact on financial market microstructure, and to those of the underlying index and provides an
questions of investor suitability. overview of the mechanics of the implied hedging
demands resulting from the daily re-leveraging of
Specifically, leveraged ETFs must re-balance their these products. Section 3 explains the microstruc-
exposures on a daily basis to produce the promised ture implications and resulting return drag from
leveraged returns. What may seem counterintuitive trading costs associated with hedging activity. Sec-
is that irrespective of whether the ETFs are lever- tion 4 analyzes the longer-term return characteris-
aged, inverse or leveraged inverse, their re-balancing tics of these products and the value of the embedded

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 45

option within; and Section 5 summarizes our results returns.5 Our findings remain unchanged regard-
and discusses their implications for public policy. less of how leveraged returns are produced, whether
by trading in physicals on margin, equity linked
notes, futures or other derivatives besides total
2 The Mechanics of leveraged returns return swaps.

2.1 Producing leveraged returns Unlike traditional ETFs, leveraged and inverse
ETFs can be viewed as pre-packaged margin prod-
As leveraged returns cannot be created out of thin ucts, albeit without any restrictions on margin
air, leveraged and inverse ETFs generally rely on eligibility. It is also worth noting that creations and
the usage of total return swaps to produce returns redemptions for leveraged and inverse ETFs are in
that are a multiple of the underlying index returns. cash, while for traditional ETFs this is typically an
Futures contracts can also be used in addition to, “in-kind” or basket transfer.
or instead of, total return swaps. However, given
their exchange-imposed standardized specification 2.2 Conceptual framework
(to facilitate exchange-based trading and clearing),
futures are not as customizable as total return swaps Now we turn to the development of a unified con-
and are more limited in terms of index representa- ceptual framework to analyze inverse and leveraged
tion. In addition, basis risk is more significant with ETFs. We will utilize a continuous time framework.
the futures than with total return swaps.3 All extant leveraged and inverse ETFs promise to
deliver a multiple of its underlying benchmark’s
Leveraged returns can also be produced by trading daily returns, so we will focus on the dynamics of
in physicals on margin. In other words, by bor- the index and of the corresponding leveraged and
rowing the required capital in excess of its AUM, inverse ETFs over a discrete number of trading days
a leveraged ETF can invest in a properly levered indexed by n where n = 0, 1, 2, . . . , N . Let tn rep-
position of the securities comprising the ETF’s resent the calendar time of day n, measured as a real
index benchmark. A negative implication of such number (in years) from day 0. We assume t0 = 0
an implementation strategy is that the financing initially, a convenient normalization. Note the fre-
cost will create a drag on the fund’s performance quency of n does not have to be daily. If there are
with respect to its promised leveraged return. On leveraged or inverse ETFs designed to produce a
the other hand, an inverse or leveraged inverse ETF multiple of the underlying benchmark’s return over
can short the securities comprising the ETF’s index a different frequency (e.g., hourly, weekly, monthly,
benchmark and accrue interest income. Interest- quarterly, etc.), then we can redefine n accordingly
ingly, a new breed of leveraged and inverse ETFs without any loss of generality.
has recently emerged that are managed against
customized index benchmarks. These benchmarks Let St represent the index level which a leveraged
explicitly incorporate the financing cost (for lever- or inverse ETF references as its underlying bench-
aged ETFs) or accrued interest (for inverse and mark at calendar time t. Later, in Section 4, we
leveraged inverse ETFs) in index construction. will explicitly describe the continuous time process
Consequently, financing cost and accrued interest underlying the evolution of the index level, but for
will not appear as a deviation against the funds’ now let rtn−1,tn represent the return of the underlying
index benchmark.4 Throughout this paper, we will index from tn−1 to tn , where
assume that leveraged and inverse ETFs rely on total Stn
rtn−1 ,tn = −1 (1)
return swaps to produce the promised leveraged Stn−1

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

46 MINDER CHENG AND ANANTH MADHAVAN

We will assume there are no dividends through- then goes up 10% on the subsequent day. Over
out to focus on the price and return dynamics the two-day period, the index declines by −1%
without any loss of generality. Let x represent (down to 90, and then climbing to 99). While an
the leveraged multiple of a leveraged or inverse investor might expect the leveraged fund to decline
ETF. Therefore x = −2, −1, 2 and 3 correspond by twice as much, or −2%, over the two-day period,
to double-inverse, inverse, double-leveraged and it actually declines further by −4%. Why? Dou-
triple-leveraged ETFs. bling the index’s 10% fall on the first day pushes
the fund’s NAV to $80. The next day, the fund’s
NAV climbs to $96 upon doubling the index’s 10%
2.3 Return divergence and path dependency
gain. This example illustrates the path dependency
Later it will become clear that the exposures of total of leveraged ETF returns, a topic we return to
return swaps underpinning leveraged and inverse more formally when we model the continuous time
ETFs need to be re-balanced or re-set daily in order evolution of asset prices in Section 4.
to produce the promised leveraged returns. In effect,
these funds are designed to replicate a multiple of 2.3.1 Example: DUG and DIG
the underlying index’s return on a daily basis. The
Real world examples of the effects noted above—
compounding of these daily leveraged moves can
and the confusion they cause among retail
result in longer-term returns, as expressed by:
investors—are not difficult to find. The relation


N between short- and long-run performance of lever-
(1 + xrtn−1 ,tn ) (2) aged ETFs is well illustrated in the case of the −2×
n=1 ProShares UltraShort Oil & Gas (DUG) and its
2× long ProShares counterpart (DIG) that track
that have a very different relationship to the longer-
the daily performance of the Dow Jones US Oil &
term returns of the underlying index leveraged
Gas index. As shown in Figure 1, these funds
statically, as given by:
are mirror images of each other over short peri-
(1 + xrt0 ,tN ) (3) ods of time, in this case a few trading days are
in March. Over longer periods, however, the per-
We can use a double-leveraged ETF (x = 2) with formance is materially different in the six month
an initial NAV of $100 as an example. It tracks an period as shown in Figure 2. Indeed between
index that starts at 100, falls 10% on day 1 and September 2008 and February 2009, both ETFs

Figure 1 DUG versus DIG (March 2–6, 2009)

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 47

Figure 2 DUG versus DIG (September 2008–March 2009).

were down substantially. These examples illustrate At the same time, reflecting the gain or loss that is x
the path-dependency highlighted in the analysis. times the index’s performance between tn and tn+1 ,
the leveraged fund’s NAV at the close of day n + 1
2.4 Re-balancing and hedging demands becomes:
The re-balancing of inverse and leveraged funds Atn+1 = Atn (1 + xrtn ,tn +1 ) (7)
implies certain hedging demands. Since extant
funds promise a multiple of the day’s return, which suggests that the notional amount of the total
it makes sense to focus on end-of-day hedging return swaps is required before the market opens on
demands. One benefit of modeling returns in the next day to maintain constant exposure is:
continuous time, however, is that our analysis Ltn+1 = x Atn+1 (8)
generalizes to any arbitrary re-balancing interval.
Indeed, there has been recent discussion of new = x Atn (1 + x rtn ,tn+1 ) (9)
leveraged funds that track an index on a monthly The difference between Eqs. (6) and (9), denoted
versus daily basis. by
tn+1 , is the amount by which the exposure of
Let Atn represent a leveraged or inverse ETF’s NAV the total return swaps that need to be adjusted or
at the close of day n or at time tn . Corresponding re-hedged at time tn+1 , as given by:
to Atn , let Ltn represent the notional amount of the
tn+1 = Ltn+1 − Etn+1 (10)
total return swaps exposure that is required before
= Atn (x 2 − x)rtn ,tn+1 (11)
the market opens on the next day to replicate the
intended leveraged return of the index for the fund
from calendar time tn to time tn+1 . With the fund’s 2.4.1 Example: daily hedging demands
NAV at Atn at time tn , the notional amount of the
total return swaps required is given by: We can illustrate the above using an example that
is built on the same case already discussed in Sec-
Ltn = x Atn (4) tion 2.3. With an initial NAV of $100 on day 0
On day n + 1, the underlying index generates a for the double-leveraged ETF, the required notional
return of rtn ,tn+1 and the exposure of the total return amount of the total return swaps is $200 (or 2 times
swaps, denoted by Etn +1 , becomes: $100). As the index falls from 100 to 90 on day 1,
the fund’s NAV drops to $80 whereas the exposure
Etn+1 = Ltn (1 + rtn , tn+1 ) (5) of the total return swaps falls to $180, reflecting
= x Atn (1 + rtn , tn+1 ) (6) a 10% drop of its value. Meanwhile the required

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

48 MINDER CHENG AND ANANTH MADHAVAN

notional amount for the total return swaps for day the funds are not leveraged or inverse), the reset or
2 is $160 (or 2 times $80), which means the fund re-balance flows are always in the same direction as
will need to reduce its exposure of total return swaps the underlying index’s performance. So, when the
by $20 (or $180 minus $160) at the end of day 1. underlying index is up, additional total return swaps
And note 100×(22 −2)×10% = 20. The dynam- exposure must be added, but when the underlying
ics of Stn , Atn , Ltn , Etn and
tn for n = 0, 1 and 2 index is down, the exposure of total return swaps
are summarized in Table 1. needs to be reduced. This is intuitively clear for
a leveraged long fund, is always true whether the
Tables 2 and 3 provide examples for an inverse ETFs are leveraged, inverse or leveraged inverse.
ETF and a double-inverse ETF, respectively, with Why is the effect the same for funds that are short
the same assumptions of the index’s performance the index? Intuitively, an inverse or leveraged inverse
over two days. These examples highlight the criti- fund’s NAV will increase if the index falls, which
cal role of the hedging term (x 2 − x) in Eq. (11). requires it to increase short exposure still further,
This term is non-linear and asymmetric. For exam- generating selling pressure. In other words, there
ple, it takes the value 6 for triple-leveraged (x = 3) is no offset or “pairing off ” of leveraged long and
and double-inverse (x = −2) ETFs. As (x 2 − x) short ETFs on the same index, which is why the
is always positive (except for when x = 1 when re-balance flows are in the same direction between
Table 1 for a double-leveraged ETF and Tables 2
Table 1 Dynamics of a double-leveraged ETF (x = and 3 for an inverse and a double-inverse ETF on
2, rt0 ,t1 = −10% and rt1 ,t2 = 10%). day 1 and day 2, respectively.
Stn Atn Ltn Etn
tn Note that the need for daily re-hedging is unique
n=0 100 100 200 — — to leveraged and inverse ETFs due to their product
n=1 90 80 160 180 −20 design. Traditional ETFs that are not leveraged or
n=2 99 96 192 176 +16 inverse, whether they are holding physicals, total
return swaps or other derivatives, have no need to
re-balance daily. We discuss the implications of daily
Table 2 Dynamics of an inverse ETF (x = −1, rt0 ,t1 = re-balances in the next section.
−10% and rt1 ,t2 = 10%).
Stn Atn Ltn Etn
tn 3 Market structure effects
n=0 100 100 −100 — —
n=1 90 110 −110 −90 −20
3.1 Liquidity and volatility near market close
n=2 99 99 −99 −121 +22 As leveraged and inverse ETFs gather more assets,
the impact of their daily re-leveraging on public
equity markets is raising concerns. Many com-
Table 3 Dynamics of a double-inverse ETF (x = mentators have cited leveraged and inverse ETF
−2, rt0 ,t1 = −10% and rt1 ,t2 = 10%). re-balancing activity as a factor behind increased
Stn Atn Ltn Etn
tn volatility at the close.6 Of course, other factors
might also account for heightened volatility at the
n=0 100 100 −200 — —
close. For example, traders might choose to trade
n=1 90 120 −240 −180 −60
later in the day because there are more macroeco-
n=2 99 96 −192 −264 +72
nomic announcements earlier in the day or because

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 49

the price discovery process takes longer. Neverthe- To understand this issue, we can extend the above
less, as we show in this section, there are good conceptual framework to model the stochastic
theoretical and empirical grounds to support the dynamics of the market impact induced by the
argument that daily re-leveraging by leveraged and hedging demand or daily re-balancing flows.
inverse ETFs contributes to volatility.
In the canonical microstructure framework, the
In theory, re-balancing activity should be exe- price movement over a trading interval is modeled as
cuted as near the market close as possible given a function of net order flow, capturing both inven-
the dependence of the re-balancing amount on tory and information effects. At the end of day n,
the close-to-close return of the underlying index the hedging demand of leveraged and inverse ETFs
as per Eq. (11). Whether leveraged and inverse is almost fully predictable, and hence it is reason-
ETFs re-balance their exposure of total return swaps able to assume that this has a same-day effect on the
immediately before or after the market close, how- close. We assume a portion 0 ≤ φ ≤ 1 of the ETF
ever, their counterparties with which they execute hedging flow contributes to price impact on day n
total return swaps will very likely put on or adjust (at the close) and 1 − φ spills over to the following
their hedges while the market is still open to min- day. So, if φ = 1, the entire impact of the hedging
imize the risk to their capital and position taking. demand is realized at the close; if φ = 0, the price
So, as leveraged and inverse ETFs gain assets, there effect is fully realized only on the successive day.
likely will be a heightened impact on the liquid-
We model the price movement as:
ity and volatility of the underlying index and the
securities comprising the index during the closing Stn+1 − Stn = λ(qtn ,tn+1 + φ
tn
period (e.g., the last hour or half-hour) of the day’s + (1 − φ)
tn−1 ) + wtn ,tn+1 (12)
trading session.
Here λ > 0 is the price impact coefficient (which in
The magnitude of the potential impact from turn can be thought of as an increasing function of
re-balancing activity is proportional to the amount volatility and decreasing function of average daily
of assets gathered by leveraged and inverse ETFs in volume), and qtn ,tn+1 is the signed order flow from
aggregate, the leveraged multiple promised by these market participants, excluding the hedging demand
funds, and the underlying index’s daily return. Such from leveraged and inverse ETFs. The stochastic
predictable and concentrated trading activity near error term wtn ,tn+1 captures the effect of unpre-
the market close may also create an environment dictable news shocks or noise trading. The hedging
for “front running” and market manipulation.7 demand itself depends on the day’s price change:
Merely substituting total return swaps with alter-
native instruments such as trading in the physicals
tn+1 = atn (x 2 − x)(Stn+1 − Stn ) (13)
on margin, futures, equity-linked notes or other
where atn = Atn /Stn . Substituting and rearranging,
derivatives will have the same economic impact
we get
as long as the product design is based on taking
leveraged positions in these instruments. Stn+1 − Stn
λ(qtn ,tn+1 + (1 − φ)
tn−1 ) + wtn ,tn+1
=
3.2 Order flow and prices 1 − λφatn (x 2 − x)
Stochastic variation in hedging demands will influ- (14)
ence volatility and return dynamics of the underly- where we require that λφatn (x 2 − x) < 1 for
ing index and the securities comprising the index. equilibrium to be defined. This expression makes

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

50 MINDER CHENG AND ANANTH MADHAVAN

clear the magnified impact of hedging demands on sub-indexes/sectors including Financials, Oil &
overall price movements by increasing the market Gas, Real Estate, Materials, etc. This sample is
impact coefficient for all flows, irrespective of their largely the universe of domestic equity leveraged
source. Intuitively, hedging demand provides an funds, excluding commodity and currency lever-
additional momentum effect to same-day returns aged funds. Ranked by notional value (as of January
that increases the price pressure effect of any signed 2009) the top three ETFs are the ProShares Ultra
order imbalance regardless of its source. S&P 500 (SSO), UltraShort S&P 500 (SDS),
and the ProShares Ultra Financials (UYG) with
The price impact is greater with higher volatility,
notional AUM of $3.0 billion, $2.9 billion, and
lower market liquidity, higher same-day effects,
$1.7 billion, respectively. Also included are the
increased AUM, and higher leverage ratios. The
new 3× and −3× funds from Direxion cover-
presence of the lagged hedge term
tn−1 (when φ <
ing the Russell 1000 (BGU and BGZ) and other
1) induces serial correlation in returns because the
indexes.
previous period’s hedge is linearly related to the pre-
vious return. Thus, the analysis shows that hedging Figure 3 shows the percentage of AUM for the uni-
flows in leveraged and inverse ETFs can exacerbate verse of US equity leveraged and inverse ETFs, bro-
same-day volatility, add momentum effects, and ken down by leverage factor (x = −3×, . . . , 3×)
induce serial correlation in returns. for end-January, 2009. As of end-January 2009, the
sample comprises 84 equity leveraged and inverse
3.3 Aggregate hedging demand ETFs with a total of $19 billion in AUM, of which
$1.4 billion or 7.3% has x = 3 or −3, a category
Recent volatility highlights the effect of hedging that did not exist in 2007.
demands at the close. How large are these effects?
To simulate the aggregate impact, we examined 83 For the above same 84 US equity leveraged and
leveraged and inverse ETFs on various US equity inverse ETFs, Table 4 reports details on vari-
indexes including S&P 500, 400, and 600 indexes, ous statistics of interest by leverage factor (x =
Nasdaq QQQ, Russell 1000, 2000, and various −3×, . . . , 3×) as of January 2009.

Figure 3 AUM in US Equity Leveraged and Inverse ETFs by Leverage Factor (x)—January 2009.

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 51

Table 4 Summary statistics on US equity leveraged For each ETF, indexed by i = 1, . . . , K , we com-
and inverse ETFs (January, 2009). pute the notional hedge as before and hence the
Leverage factor (x) −3 −2 −1 2 3 aggregate value of hedging demand at the end of day
n can be expressed as a function of the hypothetical
Bid-Ask Spread 10.5 35.3 11.8 42.4 16.8 return as
tn−1 (rtn−1 ,tn ) where:
(basis points) K 
At/Within 64.3 81.5 81.5 88.0 68.5 
Quotes (%)
tn−1 (rtn−1 ,tn ) = Ai , tn−1 (xi2 − xi ) rtn−1 ,tn
Depth ($ ’000s) 45.7 249.2 87.2 134.2 33.3 i=1
Order Size 49.3 67.8 23.5 17.8 23.7 (15)
($ ’000s) where Ai , tn−1 is the AUM of ETF i at the close of
Expense Ratio 81.7 89.4 95.0 89.4 81.7 day n − 1 or at time tn−1 .
(basis points) The Table 5 shows the function
tn−1 (rtn−1 , tn ) (in
Holding Period 1.2 6.7 8.6 15.2 1.7 millions of dollars) for returns to the correspond-
(days) ing stock indexes from 1% to +15% as of end
February 2009. We assume that all sub-indexes had
the same return, which is an obvious simplification
Several points are worth noting. Although bid- but not unreasonable given the highly correlated
ask spreads are reasonably small, their economic nature of the current market. Also shown is the
impact is large given the short holding periods. hedging demand as a percentage of median market-
Most transactions occur at or within the quotes. on-close volume. We use the median as volume is
Liquidity is generally good, especially in double- heavily skewed by end-of-month flows, quarterly re-
leveraged products, where the ratio of depth (i.e., balances, and option expiration cycles, but the result
mean quote size in dollars) to order size (in dollars) is is very similar if the mean were used instead. Note
high. that since we are essentially computing a weighted
average, the market-on-close demand in value terms
Management fees and other costs are large relative to is just a linear function of return. However, the
most traditional ETFs and do not vary significantly demand as a fraction of closing volume is a non-
by leverage. Average holding periods (calculated as linear function of return because the re-balancing
the ratio of month-end shares outstanding to aver- demand is added to the denominator too. So, for
age daily share volume for the month) are shortest example, a broad 1% move in the US equity market
for the most levered products, but are still signifi- would result in additional MOC demand of about
cant for double-leveraged ETFs (15.2 days) which 17%, while a 5% move is associated with demands
account for the bulk of AUM. Some of the most equal to 50% of the close.
recent and active leveraged funds (e.g., Direxion As AUM changes and more levered and ultra short
Daily Financials −3× Bear (FAZ)) have holding products gain traction, we would expect this slope
periods as short as 0.2 of a trading day. Note that
these are averages over heterogeneous investor pop-
Table 5
tn−1 (rtn−1 ,tn ) in $ million.
ulations. It is quite likely that holding periods vary
by investor category. For example, sophisticated rtn−1 ,tn 1% 5% 10% 15%
hedge funds may employ leveraged ETFs for within

tn−l (rtn−1 ,tn ) 787 3.934 7.873 11.811
day bets while retail investors may have longer
% MOC volume 16.8% 50.2% 66.9% 75.2%
holding periods.

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

52 MINDER CHENG AND ANANTH MADHAVAN

to steepen. Indeed, the amount of hedging has been contribution to the day’s return of the closing
increasing. Using each ETF’s AUM for December period. The coefficient β2 captures the additional
31, 2008 and December 31, 2007, we compute impact of re-balancing flows on market movements
the end-of-day dollar flows for a 5% return. This in the closing period over and above the daily return
yields end-of-day flows of $3.5 and $2.1 billion, movement.
respectively. Observe that the hedging demand cor-
The regression estimates yield β1 = 0.249 and
responding to a 5% move has already risen to $3.9
β2 = 0.541 with t-statistics of 5.806 and 2.503,
billion, almost double from year-end 2007. These
respectively, and R 2 = 0.35 with 825 degrees of
are likely conservative estimates as we exclude re-
freedom. The coefficient β0 is economically and sta-
balancing flows from leveraged and inverse mutual
tistically insignificant (0.01) as we would expect in
funds.
a model of signed returns without directional bias.
The closing period is 15.4% of the total trading day,
3.4 Regression analysis of closing period volatility
but is a higher fraction (about 25%) of the day’s
Now we turn to a more direct test of the pre- volume and hence of the day’s return as implied by
diction that re-balancing flows exacerbate return the estimates. The estimate of β2 suggests that after
movements towards the close. As leveraged funds controlling for the day’s return movement, $1 bil-
typically will re-balance towards the end of the lion of re-balancing flow adds 0.54% to the closing
day, we focus on returns in last hour of the trad- period’s return.
ing day for large capitalization stocks. Specifically,
Since concerns about re-balancing flows are often
we look at re-balancing flows arising from 2× and
framed in terms of volatility, we also examine the
3× leveraged and inverse funds (SSO, SDS, BGU,
second moment of closing period returns. Specifi-
and BGZ) on large capitalization indexes. The ETF
cally, we want to see if there is an empirical relation
aggregate re-balancing flows are then computed as
between absolute returns in the closing period and
above for each day. This is likely a conservative esti-
the magnitude of re-balancing flows. It is possible
mate because there is also large cap re-balancing flow
that more volatile days will see greater movements
from other leveraged and inverse ETFs (e.g., SKF)
at the close, so we use the daily value of the VIX
on narrower benchmark indexes that contain large
index as a control variable. Accordingly, we run the
cap stocks.
following regression using daily data from January
We model the closing period returns as a function 1, 2006 to mid-April 2009:
of the day’s return and signed re-balancing flows.
Using daily data over the period January 1, 2006 |rtn− ,tn | = βo + β1 Vtn + β2 |
tn | (17)
to mid-April 2009, we then estimate the following where on day n, |rtn− ,tn | is the absolute value of the
regression equation: closing period return (defined as the intra-day S&P
rt − ,tn = β0 + β1 rtn−1 ,tn + β2
tn (16) 500 return from 3 p.m. Eastern time to the close),
Vtn is the value of VIX at the close of day n, and
where tn− denotes the calendar time of 3 p.m. East-
|
tn | is the absolute value of aggregate re-balance
ern time on day n, rtn− ,tn is the intra-day S&P 500
flow (in billions of dollars).
return from 3 p.m. Eastern time to the close on
day n, rtn−1 ,tn is the close to close S&P 500 return, We estimate β1 = 0.023 and β2 = 0.724 with
and
tn is the signed aggregate re-balance flow (in t-statistics of 12.186 and 6.422, respectively. The
billions of dollars). Equation (16) has a straight- coefficient β0 is small in value, −0.18, with a t-
forward economic interpretation: β1 captures the statistics of −4.526. Model fit is good; R 2 = 0.457

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 53

with 825 degrees of freedom. The coefficient on β1 Since Eq. (20) holds for any period, when the time
is no surprise; higher volatility means greater return interval between tn and tn+1 is sufficiently small,
movements in the closing period. Controlling for this expression becomes:
overall volatility, the significantly positive coeffi- dAt dSt
cient β2 confirms that greater imbalances add to =x (21)
At St
closing volatility. These results also hold with other
measures of closing volatility or with the dependent where At and St represent the NAV of a leveraged or
variable expressed as a fraction of VIX. Clearly, the inverse ETF and the ETF’s underlying index level,
analysis above can be extended in many ways includ- respectively, at time t. Note that as described earlier,
ing the examination of other indexes and countries, a leveraged ETF hedges at discrete time intervals
but this lies outside the scope of this paper. but always maintains economic exposure at x× the
underlying index return, as represented in Eq. (21).
With Eqs. (21) and (18), it can be shown that:
4 Return dynamics
dAt = xµAt dt + xσAt dWt (22)
Having discussed the mechanics of leveraged and
inverse ETFs and market structure implications of and then ln(At ) follows a generalized Wiener pro-
daily re-balancing activity, we turn now to an anal- cess, as given by:
ysis of returns over longer periods of time. Our  
x 2σ 2
objective is to understand return dynamics and the d ln(At ) = xµ − dt + xσdWt (23)
role of the embedded option within the leveraged 2
and inverse product to address questions concern- suggesting that At follows a geometric Brownian
ing the suitability of these products for individual motion with a drift rate of xµ and a volatility
investors, particularly those with longer holding of xσ. In other words, the volatility of a lever-
periods. To do this, we now need to model explicitly aged or inverse ETF is just x times the volatility
the evolution of security prices in continuous time. of its underlying index. This is handy for pric-
Specifically, the index level St is assumed to follow ing options on leveraged and inverse ETFs. Such
a geometric Brownian motion, as given by: options are gaining popularity and volume in par-
allel with activity in their underlying ETFs, and
dSt = µSt dt + σSt dWt (18) there are now over 150 different option contracts
traded on these instruments.
with a drift rate of µ and a volatility of σ. Here,
Wt in Eq. (18) is a Wiener process with a mean
of zero and a variance of t. Then ln(St ) follows a 4.1 Leveraged ETFs and the underlying index
generalized Wiener process, as given by: The various relationships derived in a continuous-
  time setting, i.e., Eqs. (21)–(23), now can be
σ2
d ln(St ) = µ − dt + σdWt (19) applied to our earlier analysis at the daily level.
2
Applying Eq. (19) to Stn and S0 gives us the fol-
We need to relate the dynamics of the NAV of a lowing relationship of the index level on day N and
leveraged or inverse ETF to the dynamics of the day 0, or between time tN and time 0:
ETF’s underlying index level. From Eq. (7), it can  σ2 √
be shown algebraically that: StN = S0 exp µ − tN + σ tN z (24)
2
 
Atn+1 − Atn Stn+1 − Stn where exp(z) = e z is the exponential function and
=x (20)
Atn Stn z is a standard normal distribution with a mean of 0

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

54 MINDER CHENG AND ANANTH MADHAVAN

and a standard deviation of 1. So, the return to the 4.2 Embedded option
index over the period, ln(StN )− ln(S0 ), is normally
Equation (27) also suggests that the NAV of a lever-
distributed with mean [µ − σ2 ]tN and standard
2

√ aged or inverse ETF will always be positive.8 This

deviation σ tN . Likewise, applying Eq. (23) to
is in contrast to the risk implied by Eq. (3) that
Atn and A0 will give us the following relationship of
a statically leveraged fund may see its NAV com-
the leveraged or inverse ETF’s NAV on day N and
pletely wiped out, for example, if over a N -day
day 0:
period its underlying index has doubled for a short
  fund, or the index has lost half of its value for a
x 2 σ2 √
AtN = A0 exp xµ − tN + xσ tN z double-leveraged fund. The preservation of cap-
2
(25) ital is made possible because, unlike a statically
leveraged fund which establishes its target leveraged
In other words, the return to the leveraged notional amount initially and keeps it unchanged,
ETF ( ln(AtN ) − ln(A0 )) is normally distributed a leveraged or inverse ETF dynamically adjusts its
with mean [xµ − x 2σ ]tN and standard deviation
2 2
leveraged notional amount of total return swaps on
√ a daily basis depending on the performance of its
xσ tN . These draws are not independent, however,
because the same realization of the sample path, underlying index. In effect, leveraged and inverse
captured by z, is in both returns. ETFs have an embedded option on the underlying
index.9
We can further simplify this as follows:
Given the dynamic nature of the leveraged and

xσ 2 √ inverse ETFs’ daily re-balancing activity, the value
AtN = A0 exp xµ − tN + xσ tN z of the embedded option is dependent on not only
2
 StN but also the entire path of the underlying index
(x − x 2 )σ 2 tN prior to reaching StN at time tN . Table 6 provides
+ (26)
2 a simple example. It is based on the same double-
 x   leveraged ETF as in Table 1. As in Table 1, the index
StN (x − x 2 )σ 2 tN
= A0 exp (27) ends at 99 on day 2, however, it takes a different
S0 2 path in Table 6 to get there: The index is flat on day
1 and down 1% on day 2. Note that the NAVs on
So, the total return of a leveraged or inverse ETF
day 2 (At2 ) are different between Tables 6 and 1.
over a N -day period (i.e., (AtN /A0 )) is the xth power
of the gross return of the ETF’s underlying index The effect of such volatility is captured in the scalar
over the same time period (StN /S0 ), multiplied by 2 2
2 2 of exp( (x−x 2)σ tN ) in Eq. (27). As (x − x 2 ) is neg-
a scalar exp( (x−x 2)σ tN ). As (x − x 2 ) < 0, the scalar ative for the leveraged and inverse ETFs, the scalar
is less than 1. As shown in Section 2, leveraged
and inverse ETFs are designed to replicate a mul-
tiple of their underlying index’s return on a daily Table 6 Dynamics of a double-leveraged ETF (x =
basis. As it turns out, ignoring the scalar in Eq. (27) 2, rt0 ,t1 = 0% and rt1 ,t2 = −1%).
the compounding of these daily leveraged moves is Stn Atn Ltn Etn
tn
economically equivalent to compounding the gross
return of the index to the xth power. Note that the n=0 100 100 200 – –
risk free rate does not enter into Eq. (27) because n=1 100 100 200 200 0
this is implicit in the costs of the total return swaps. n=2 99 98 196 198 −2

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 55

is always less than 1 and an ideal path is one with raised to the xth power, then multiplied by the scalar
2 2
lower volatility (σ) and shorter time horizon (tN ). exp( (x−x 2)σ tN ). This scalar gives rise to the value of
For example, as the index has demonstrated a sig- the embedded option in the first place, but is also
nificantly lower volatility over the two-day period the source—under certain conditions—of long-run
in Table 6 than in Table 1, note the NAV on day 2 erosion in value.
for the double-leveraged ETF in Table 6 is higher
than that in Table 1 ($98 vs. $96). To see this, note that the scalar decreases with tN and
asymptotically approaches zero as tN approaches
Of course, in a practical context, we need to con- infinity. Along any sample path of returns, the gross
sider additional factors. First, actual payoffs are return of the leveraged or inverse ETF (relative to
going to depend on the realized sample paths of the the underlying index return) will decline with the
random variables making up the N -period returns. duration of holding and the volatility of the under-
Second, we need to distinguish between calendar lying index. This problem is exacerbated for higher
and trade days. Third, these expressions assume the leveraged (i.e., x) ETFs. In other words, the longer
management and other fees for the leveraged and the buy-and-hold period of a leveraged or inverse
inverse ETFs is the same as their un-levered coun- ETF, the lower the ETF’s long-term return relative
terpart, whereas they are typically much higher. to x times the underlying index return. For example,
Further, as shown above, leveraged, inverse, and holding a triple-leveraged (x = 3) or double-inverse
leveraged inverse ETFs are designed to replicate a (x = −2) ETF for five years (tN = 5) when the
leveraged exposure to the underlying index on any volatility of the index returns is 50% (σ = 50%),
one day, but not necessarily over extended periods the scalar becomes 2.3%.
of time. A leveraged or inverse ETF’s return beyond
a day is determined by the index returns and volatil- Of course, the total return to a leveraged or inverse
ity, leverage (financing) costs, expenses, and the ETF along any sample path could greatly exceed
time horizon. The inability of leveraged and inverse the underlying index return even if the holding
ETFs to track index returns over periods longer than period is relatively long. Note that the first term in
a day is not widely understood by retail investors Eq. (27) reflects the realized sample path of index
in these funds. Note: The leveraged ETF’s man- returns. If the mean return is large and positive and
ager must re-invest at the close of each day without x > 0, then this term can imply substantial pay-
actually observing the close, which induces further offs to the leveraged ETF holder over many sample
slippage. realizations. Consequently, it is of interest to exam-
ine the expected value of a longer-term investment.
This expression can be explicitly derived using the
4.3 Value destruction of buy-and-hold moment generating function of the lognormal dis-
tribution. Specifically, if Z is a random variable that
This above analysis helps us to understand questions is distributed normally with mean m and standard
of investor suitability. While leveraged and inverse deviation s, it can be shown that the expected value
ETFs do offer limited liability, this comes at a cost. of exp(kZ ) is
It was mentioned before that unlike a statically
 
hedged fund, leveraged and inverse ETFs contain an k2s2
embedded option embedded in the product design. E [ exp(kZ )] = exp km + . (28)
2
Recall from Eq. (27) that the N -period realized
gross return of a leveraged or inverse ETF is just the Recall the N -period index gross return ln(StN ) −
ln(S0 ) has a mean [µ − σ2 ]tN while the
2
gross return of its underlying index over that period

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

56 MINDER CHENG AND ANANTH MADHAVAN

corresponding x-times leveraged or inverse ETF return:

has a mean [µx − x 2σ ]tN . Then, using Eq. (28)
2 2
1. x < 0 (e.g., inverse and leveraged inverse ETFs)
with m equal to the corresponding mean, the and µ > 0, i.e., the index drift is positive;
expected gross return of the index at time tN is 2. x > 0 (e.g., leveraged long ETFs) and 0 < µ <
exp([µ − σ2 ]tN + σ 2tN ) = exp(µtN ), while the
2 2
xσ 2
2 ; and
expected value of the leveraged or inverse ETF is 3. x > 0 and µ < 0, i.e., negative drift of index
exp(xµtN ). Intuitively, this result is a reflection of returns.
Jensen’s inequality and the convexity of the expo-
nential function. It follows immediately that if In case 1, the returns to inverse ETF holders (i.e.,
xµ < 0 (i.e., an inverse ETF with positive index x < 0) over long periods of time should be negative
drift or a leveraged ETF with negative index drift) because we expect µ > 0 in equilibrium so the
that the expected value of a leveraged or inverse ETF first term in Eq. (27) will also tend to zero as tN
will tend to zero, i.e., a scenario of long-term value increases. In other words, the long-run inverse and
destruction. leveraged inverse ETF value is zero. In case 2, the
variance term dominates µ so the leveraged ETF
It is important to understand that the expected
has a negative drift. An interesting special case is
value derived above is a mean value; it is affected
where the index return has positive median returns
by the small probabilistic weight placed on extreme
but the corresponding leveraged ETF is negative.
sample paths. Because the lognormal distribution
Then, for x > 0, if µ satisfies: σ2 < µ < xσ 2 , the
2
is positively skewed, the simple expectation is not
likely the typical experience of the average investor. long-run return of the leveraged or inverse ETF is
Consider a simple numerical example: suppose negative despite a positive drift in the index. Case
index returns are equally likely to be 4% or −2%. 3—the polar opposite to case 1—seems unlikely
Over a successive two-day period, the four possible in a long-run equilibrium, but it is included for
cumulative returns are 8.16%, 1.92%, 1.92%, and completeness.10
−3.96%. The average two-day cumulative return In summary, theory shows that leveraged and
is thus 2.01%, but only one sample path (i.e., suc- inverse ETFs are generally not suitable for buy-
cessive daily 4% returns) of the four exceeds this and-hold investors.11 Indeed, it is possible for the
figure. Since the skewness of the lognormal distri- median investor to face a negative return drift in
bution increases with variance, the mean value is a leveraged or inverse ETF although the underly-
especially unrepresentative for volatile assets. Con- ing index has a positive drift. But while levered
sequently, it makes sense to focus on the typical products can yield less than x times the index
or median investor’s value change over an interval return, their volatility is exactly x times that of
of time. Again, from the properties of the lognor- the index. From a long-run equilibrium perspec-
mal distribution, if Z is distributed normally with tive, the performance drag is most problematic for
mean m and standard deviation s, the median value inverse ETFs and products based on volatile under-
of exp(Z ) is exp(m), which does not depend on the lying indexes, as the return scalar also decreases
variance. with σ.

Formally, substituting m = (xµ − x 2σ )tN for the

2 2

4.4 Frictions
mean return in a given interval of time tN , we
can now explicitly characterize situations in which From a practical standpoint, the cumulative impact
leveraged or inverse ETFs have a negative median of transaction costs can dramatically affect the

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 57

returns to a long-run investor in these ETFs. It At the beginning of each day, the fund pays a
is useful to distinguish between explicit costs (e.g., dealer a total transaction cost Ctn proportional to
management fees) and implicit costs (e.g., dealer the dollar amount of the total notional swap expo-
hedging costs and the impact costs arising from daily sure. Let θ denote the hedging cost (dealer’s fee)
re-balancing activity) which represent a further drag as a constant fraction of the notional amount of
on longer-term performance. total return swaps required, where 0 ≤ θ < 1,
so that Ctn = θEtn . Net of costs, the exposure
Explicit costs are easy to account for as they are a in Eq. (6) is just scaled by (1 − θ), i.e., Etn+1 =
constant fraction of the fund’s AUM. Total expense (1 − θ)xAtn (1 + rtn ,tn+1 ). Note that the risk free
ratios for leveraged and inverse ETFs range from rate is implicit in θ, although it is negligible at
75–95 basis points, which are high relative to the daily level. The total notional swap exposure
traditional ETFs. But while explicit costs are trans- is directly reduced by transaction costs (much like
parent, the hidden performance drag caused by a tax) so that the gain or loss is (1 − θ)x times
implicit costs is not readily quantified and is not the index’s performance between tn and tn+1 . So,
well understood by investors. Eq. (7) becomes
Our previous discussion in Section 2.4 and exam-
ples provided in Tables 1, 2 and 3 did not factor Atn+1 = Atn (1 + (1 − θ)xrtn ,tn+1 ) (29)
transaction cost. But as leveraged and inverse ETFs
re-balance their exposure of total return swaps on If we define x  = (1 − θ)x, we see that the analysis
a daily basis, they will incur transaction costs and of Section 4 goes through fully by merely substi-
the cumulative effect of the transaction costs on tuting x  for x throughout. Since |x  | < |x|, dealer
the ETFs’ performance cannot be under-estimated transaction costs act as a performance drag on the
given the daily nature of the funds’ re-balancing fund, reducing the promised returns.12
activity. In particular, these leveraged and inverse Daily fund re-balancing also imposes a further per-
ETFs should expect to pay fully or at least partially formance shortfall from what is promised, as given
the bid-ask spread of adjusting their total return by Eq. (2), due to market impact costs. We assume
swaps exposures. In addition, there may be market the market impact cost is proportional to the size
impact cost if the re-balancing amount on a given of the re-balancing trade each period. The expected
day demands more liquidity than the market is able market impact cost, denoted by Mtn , is modeled
to absorb in an orderly manner near the close. As as proportional to the absolute hedging demand
mentioned in Section 3.1, the leveraged and inverse |
tn |:
ETFs are always buying when the underlying index
is up, and selling when the index is down. They Mtn = E [λ(x 2 − x)|rtn−1 ,tn |] (30)
will not only incur higher transaction cost because
they almost always demand liquidity when they re- where λ > 0 is the market impact coefficient.
balance daily, but also likely create a performance Define by τ = tn − tn−1 the fixed re-balancing
drag in the long run. interval, typically a day. From our previous results,
the daily return rtn−1 ,tn is√ distributed normally
We can model the impact of implicit transaction with standard deviation σ τ. Assuming for con-
costs within the framework already developed. Let venience that the daily mean return is negligible
us begin with the costs of establishing the swap posi- and using the results for the folded normal distri-
tion. These hedging costs are visible to the fund bution (i.e., the distribution of the absolute value
(e.g., dealer spreads, etc.) but not to its investors. of a normal variate), the expected daily impact

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

58 MINDER CHENG AND ANANTH MADHAVAN

cost is 4.4.1 Example: UltraShort financials ETF

2τ
Mtn = λ(x 2 − x)σ . (31) Empirically we observe leveraged and inverse ETF
π
This cost will be most evident over many days as returns and index returns, but not the implicit costs
the cumulative impact of these costs is compounded of trading. We should, however, see a deterioration
and is manifested in the form of slippage from the in longer-term leveraged and inverse ETF returns
benchmark index return. relative to the index returns if these costs are signif-
icant. As an example, consider SKF, the ProShares
From Eq. (31), we see that market impact costs UltraShort Financials, a double-inverse ETF based
increase with volatility, σ, the market impact on the Dow Jones US Financial Services index,
parameter, λ, the leverage factor, x, and the time that has been the subject of considerable contro-
between re-balances, τ. We can use this analysis to versy. Using two years of daily data from February
understand the effects of changing the re-balancing 1, 2007 to February 1, 2009, we regressed the
frequency (e.g., to twice a day or once a week) on daily return of SKF on the daily return on IYF, the
the underlying economics. Interestingly, moving iShares Dow Jones Financial Services Index ETF.
to less frequent re-balancing can mitigate market Figure 4 shows the plot of the 502 close-to-close
impact costs. To see this, suppose we increase the returns of SKF against IYF.13 In theory, we have
re-balancing interval to τ  = 2τ, i.e., every two x = −2 but the regression has a slope of −1.94
trading days. Then, using Eq. (31), the expected re- with a very high R 2 = 0.98. As a further check,
balancing impact cost for a two-day re-balancing we find that the first order autocorrelation of both
interval is 44% higher than the daily expected products is similar (−0.07 for SKF and −0.09 for
cost, not double the daily value. Intuitively, impact IYF), indicative of small temporary price reversals.
costs are proportional to the standard deviation
√ With x  = 1.94, we estimate θ = 0.03 or 3% as
of returns, which in turn increases with τ, the our measure of transaction cost and impact. This
square-root of the calendar re-balancing interval. relatively high cost estimate is perhaps not surpris-
So less frequent re-balancing implies less absolute ing given the high volatility of the financial sector
hedging and hence lower costs for a given calen- in this period. At the monthly level over the period
dar period. It is worth emphasizing though that November 2008 to February 2009, we see a sharp
while a change in re-balancing frequency can affect deterioration in tracking ability. The slope coeffi-
implicit costs, it does not alter the economics of cient falls to −1.75 and the goodness of fit drops
the longer-run performance, as can be seen if tN is with R 2 = 0.76. This deterioration is consistent
interpreted not as days but as periods with µ and σ with the cumulative effect of market impact costs
scaled appropriately. as noted earlier. Indeed, over the period November
Taxes represent another source of friction. Leveraged 2008 to the end of March 2009, both SKF and IYF
and inverse ETFs are not designed with respect to declined by −25.1% and −36.35%, respectively,
tax efficiency and an investor’s after-tax performance as shown in Figure 5.
might be significantly lower than a strategy using
leverage with less turnover. At year-end 2008, for 5 Conclusions and policy implications
example, the major inverse ETF providers (Rydex
and ProShares) paid out substantial capital gains As noted in the introduction, the class of leveraged
tax distributions ranging from 4% to 86%. The and inverse ETFs continues to expand rapidly. The
impact of taxes, however, will vary depending on growth is especially rapid in more leveraged and
each individual investor’s particular circumstances. leveraged inverse products. From the above analysis,

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 59

Figure 4 Daily SKF and IYF returns (February 2007–2009).

Figure 5 SKF and IYF cumulative returns (November 2008–March 2009).

however, it is precisely these products that pose the order to produce the leveraged returns promised.
greatest challenges for public policy from both the Whether the ETFs are leveraged, inverse or lever-
perspective of individual investor protection and aged inverse, their re-balancing activity is always in
the larger goal of ensuring efficient and orderly the same direction as the underlying index’s daily
markets. performance: When the underlying index is up, the
additional exposure of total return swaps needs to
Specifically, leveraged and inverse ETFs generally be added; when the underlying index is down, the
rely on the usage of total return swaps to produce exposure of total return swaps needs to be reduced.
returns that are a multiple of the ETFs’ under- The cumulative effect of the transaction costs on
lying index returns. As these ETFs are in effect the leveraged and inverse ETFs’ performance can-
designed to replicate a multiple of the underlying not be underestimated given the daily nature of the
index’s return on a daily basis, their exposures of funds’ re-balancing activity. The re-balance flows are
total return swaps need to be re-balanced daily in executed as near the market close as possible given

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

60 MINDER CHENG AND ANANTH MADHAVAN

their dependence on the close-to-close return of the return swaps would alleviate some of the concerns
underlying index. The need for daily re-balancing arising from the short gamma nature of the hedging
is unique to the leveraged and inverse ETFs due to strategy. Better disclosure might also help mitigate
their product design. Traditional ETFs that are not broader counterparty credit risks. Fund companies
leveraged or inverse, whether they are holding phys- are not required to disclose the details of their swap
icals, total return swaps or other derivatives, have agreements, making it difficult to determine coun-
no such need for daily re-hedging. terparty credit quality. If counterparties are not
willing or unable to provide total return swaps, the
From a market structure perspective, many broker- leveraged ETF manager must replicate these returns
dealers believe the end-of-day flows needed to using derivatives, which can be costly and induce
maintain the leverage promised by leveraged and further tracking error.
inverse ETFs is having a heightened impact on liq-
uidity and volatility of the underlying index and Alternatives to existing products could be based
the securities comprising the index at the close. on changing the re-balancing frequency. However,
Inverse ETFs in particular have a magnified hedging shorter re-balancing (e.g., twice a day) will magnify
demand relative to their long counterparts (e.g., as intra-day price movements while extending the re-
shown above, a 2× inverse ETF has the same hedg- balancing frequency (e.g., to weekly) increases the
ing multiplier as a 3× leveraged long ETF).14 The risk of the fund being wiped out (i.e., as the fund
above analysis shows leveraged and inverse ETFs becomes gradually similar to a statically hedged
amplify the market impact of all flows, irrespec- structure). As long as product design is based on tak-
tive of source. Essentially, these products require ing leveraged positions, the underlying economics
managers to “short gamma” by trading in the same of these instruments is not altered.
direction as the market. There is a close analogy to
the role played by portfolio insurance in the crash Our results also raise questions of investor suit-
of 1987.15 The magnitude of the potential impact ability. Some investors—despite very clear language
is proportional to the amount of assets gathered in the prospectuses—do not understand the point
by these ETFs, the leveraged multiple promised, clearly that leveraged and inverse ETFs will not
and the underlying index’s daily returns. Such pre- necessarily replicate the leveraged index return over
dictable and concentrated trading activity near the periods longer than a day.16 Generally, the greater
market closes may also create an environment for the holding period and the higher the daily volatil-
front running and gaming. ity, the greater the deviation between the leveraged
ETF’s return and a statically levered position in the
These issues suggest regulators need to carefully con- same index. Also note that as leverage is embed-
sider the impact of point-in-time liquidity demands ded in the product, individual investors can gain
when authorizing leveraged products, especially additional leverage by buying these products on
leveraged inverse products or products on relatively margin. Better education, margin restrictions, and
narrow, volatile indexes. Regulators, moreover, can- tighter requirements on investor eligibility are possi-
not easily track the implicit exposure in a leveraged ble options regulators could consider. For example,
or inverse ETF (this is not the case if the under- some retail investors may be holding leveraged
lying ETF itself is levered), making it difficult to funds in accounts that are not approved for mar-
assess the market structure impact of these prod- gin, either because of lack of documentation or in
ucts as their AUM changes over time. Developing a accounts (e.g., retirement) that explicitly exclude
mechanism to track notional exposure through total leverage.

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution

 DYNAMICS OF LEVERAGED AND INVERSE ETFs 61

Our analysis of return dynamics provides condi- educational purposes and should not be construed as a
tions under which buy-and-hold investors expe- recommendation to purchase or sell, or an offer to sell or
rience eventual value destruction in leveraged or a solicitation of an offer to buy any security. We thank
Mark Coppejans, Matt Goff, Allan Lane, Hayne Leland,
inverse ETFs, a point not well understood even by
J. Parsons, Heather Pelant, Ira Shapiro, Jim Smith, Mike
experts. The gross return of a leveraged or inverse Sobel, Richard Tsai and an anonymous referee for their
ETF over a finite time period can be shown alge- helpful comments.
braically to be simply the gross return of the ETF’s 2 This figure does not include leveraged mutual funds that
underlying index over the same period raised to the have the same characteristics as leveraged ETFs except for
power of the leveraged multiple of the ETF, mul- the fact that they are not traded intradaily.
3 Basis risk refers to the risk associated with imperfect hedg-
tiplied by a scalar that is less than one. The NAV
ing, possibly arising from the differences in price, or a
of a leveraged or inverse ETF is non-negative (in mismatch in sale and expiration dates, between the asset
contrast to the risk that a statically leveraged fund to be hedged and the corresponding derivative.
may see its NAV completely wiped out) because 4 See, for example, Dow Jones STOXX Index Guide
a leveraged or inverse ETF dynamically adjusts its (2009). This structure is more common in Europe.
5 In addition to swaps, leveraged funds’ assets typically
leveraged notional amount of total return swaps
on a daily basis depending on the performance of include a pool of futures contracts to manage liquidity
demands and reduce transaction costs. Some long lever-
its underlying index. In effect, the leveraged and aged funds listed in the US are guided by a no-action letter
inverse ETFs have an embedded path-dependent from the SEC that stipulates a minimum (e.g., 80% or
option on the underlying index. The value of the more) of assets should be in the underlying. However,
option is dependent on the entire path of the under- this does not apply to the inverse funds.
6 See, e.g., Lauricella, Pulliam, and Gullapalli (2008), who
lying index. Specifically, if volatility is sufficiently
high, the median investor will experience a long-run note: “As the market grew more volatile in September,
Wall Street proprietary trading desks began piling onto
erosion in value in a leveraged or inverse ETF.
the back of the trade knowing that the end-of-day ETF-
Other considerations also might have a material related buying or selling was on its way. If the market was
falling, they would buy a short ETF and short the stocks
impact on longer-term investors. Transaction costs
or the market some other way. If the market was rallying,
in leveraged, inverse, and leveraged inverse ETFs they would buy a bull fund and go long.”
may be higher than those of replicating leveraged 7 Illegal front running refers to a broker executing orders
long or short exposure directly through swaps, for its own account before filling pending client orders
options, futures, or trading on margin. In addition, that are likely to affect the security’s price. Here, we use
the cumulative impact of transaction costs arising the term more colloquially to refer to trading ahead of
the flows that are highly predictable given public infor-
from daily re-balancing activity will only reduce
mation on index returns and fund AUM that need not be
these funds’ NAV further. There are also important illegal.
tax consequences, particularly with inverse ETFs. 8 See also Despande, Mallick, and Bhatia (2009).
In essence, the leveraged and inverse ETFs are 9 This distinction also applies to levered active products
not suitable for buy-and-hold investors. Neither and hedge funds. These funds rely on static hedges that
these products are designed to deliver long-term do not require daily re-levering. The daily re-balancing
performance for volatile indexes. activity of these funds is driven mainly by alpha changes
and risk management, and their trades do not have to be
concentrated near the market close.
10 In a general equilibrium framework, it is possible for a
Notes
risky asset to have a long-run expected return below the
1 The views expressed here are those of the authors alone. risk free rate if it acts as a hedge (e.g., with an embedded
The strategies discussed are strictly for illustrative and put), but this would seem unlikely for an equity index.

FOURTH QUARTER 2009 JOURNAL OF INVESTMENT MANAGEMENT

Not for Distribution

62 MINDER CHENG AND ANANTH MADHAVAN

11 See also Lu, Wang, and Zhang (2009) who reach a simi- Dow Jones STOXX Index Guide (2009). “Dow Jones STOXX
lar conclusion from an empirical analysis of longer term Index Guide.” STOXX and Dow Jones, Inc.
leveraged ETF returns. Grossman, S. and Vila, J.-L. (1989). “Portfolio Insurance
12 See, for example, Leland (1985) for an analysis of the in Complete Markets: A Note.” Journal of Business 62(4),
impact of volatility on expected transaction costs in the 473–476.
context of option pricing and replication. Lauricella, T., Pulliam, S. and Gullapalli, D. (2008). “Are
13 Source: Bloomberg, with closing prices and returns ETFs Driving Late-Day Turns? Leveraged Vehicles Seen
adjusted for corporate actions. Magnifying Other Bets; Last-Hour Volume Surge.” Wall
14 These arguments could also extend to the Over-the- Street Journal, December 15, C1.
Counter (OTC) counterparts of leveraged ETFs and to Leland, H. (1985). “Option Pricing and Replication with
levered non-exchange traded funds. Transactions Costs.” Journal of Finance 40(5), 1283–1301.
15 See, e.g., Grossman and Vila (1989). The working paper Lu, L., Wang, J. and Zhang, G. (2009). “Long Term Per-
underlying this article circulated prior to the Crash of formance of Leveraged ETFs.” Working paper, Baruch
October 1987. College.
16 See, e.g., Zweig (2009), who provides several examples Zweig, J. (2009). “How Managing Risk with ETFs Can
of leveraged funds failing to track the underlying index. Backfire.” Wall Street Journal, February 28, B1.
Zweig notes, however, that: “Still, many financial advisors
believe these funds are a good long-term hedge against Keywords: Leveraged and inverse ETFs; market
falling markets.” microstructure

References
Despande, M., Mallick, D. and Bhatia, R. (2009).
“Understanding Ultrashort ETFs.” Barclays Capital Special
Report.

JOURNAL OF INVESTMENT MANAGEMENT FOURTH QUARTER 2009

Not for Distribution