The Portfolio Rebalancing Effects of the ECB's Asset ... · active rebalancing behavior as distinct from passive rebalancing as described in Bubeck, Habib, and Manganelli (2017) (this

The Portfolio Rebalancing Effects of the ECB’sAsset Purchase Programme∗

Giovanna Bua and Peter G. DunneCentral Bank of Ireland

We explore the transmission of the ECB’s Public SectorAsset Purchase Programme (PSPP) via portfolio rebalanc-ing of investment funds and their investors. Using dynamicpanel methods, we find that PSPP-holding funds rebalancetowards bonds of non-EA banks and away from maturities tar-geted for purchase. Other fund types rebalance into assets withlonger durations and towards non-EA bonds issued by bothnon-financial corporations and sovereigns. Significant effectsappear after the scaling-up of the program. There is no evi-dence of rebalancing towards European equities. The patternof rebalancing over the period examined reflects a reluctanceto increase exposures to liabilities of EA banks and corporates.

JEL Codes: D72, H00, C72, C82.

1. Introduction

We extend and deepen the extant analysis of the effects of theEurosystem’s Extended Asset Purchase Programme (EAPP) byexamining the portfolio rebalancing of investment funds and, indi-rectly, the behavior of fund investors who are responsible for theinvestment flows that determine redemption and issuance of fundunits. Using a micro-panel econometric methodology, we identifyhow fund characteristics affect the heterogeneity of rebalancingbehavior during the program.

∗We thank Angelo Ranaldo, Joe McNeill, and Fergal McCann for helpfulcomments on an earlier draft. We are grateful to colleagues for their assis-tance with the funds data (including Brian Golden, Mark Bohan, SiobhanO’Connell, Padraig O’Brien, and Dermot Coates). Views expressed are thoseof the authors and do not necessarily represent the views of the CentralBank of Ireland or the ESCB. Author e-mails: [email protected];[email protected].

1

2 International Journal of Central Banking December 2019

Portfolio rebalancing is one of several channels through whichasset purchase programs are understood to transmit to economiceffects. Asset purchases cause sellers to substitute into other assetclasses. If investors have strong preferences for the holding of spe-cific asset classes (and not just maturity-based preferred habitats asin Vayanos and Vila 2009), then there could be strong asset priceeffects from actual portfolio rebalancing. This raises the prospectof more widespread beneficial outcomes such as those outlined byKrishnamurthy and Vissing-Jorgensen (2011).1

By purchasing long-term bonds, the central bank reduces theoverall duration risk borne by banks and investors.2 Whether assetpurchases have beneficial side effects (such as encouraging the takingof more risk or duration) is an open empirical question. Our analysisthrows light on the risk/duration channel by examining changes inthe term to maturity of assets held.3

Portfolio rebalancing away from euro-area (EA) assets and intoforeign-issued assets may produce beneficial effects through anexchange rate channel. Demertzis and Wolff (2016) suggest thatthere has been a strong effect from asset purchases on the valueof the euro.4 The current analysis does not attempt to identify

1A potential beneficial outcome from EAPP would be higher-priced equitymarkets which would drive down the cost of equity capital. (Analysis providedin the European Central Bank Economic Bulletin in 2015 finds little evidenceof expected improvements in earnings/dividends following the initial phase ofEAPP, so buoyant equity markets probably reflect asset purchase effects actingthrough increased risk appetite and lower discount rates).

2This also demonstrates its commitment to lower rates for longer (see, forexample, Eggertsson and Woodford 2003).

3Financial stability concerns emerge (and a risk channel arises) if the programencourages holders to switch to high-risk assets and greater leverage as in Wood-ford (2012) and Coimbra and Rey (2016), as well as in taking on more liquidityrisk as described by Stein (2014).

The theoretical work of Brunnermeier and Sannikov (2014) outlines more spe-cific mechanisms for the transmission of such risks. Valiente (2015) also mentionsthe effects of the purchases on liquidity and financial plumbing. Evidence forthese broader channels is beyond our focus.

4A January 2017 report, Deutsche Bundesbank (2017), also contributes evi-dence of such effects but concludes that they do not go significantly beyond theannouncement effects of the program. The bulletin also highlights the fact thatexchange rate effects are very difficult to disentangle from the effects of othermonetary policy actions (e.g., changes in the rate earned on funds at the deposit

Vol. 15 No. 5 The Portfolio Rebalancing Effects 3

exchange rate effects directly, but it provides evidence of significantmovements into foreign assets.

To assess portfolio rebalancing effects, we principally examinethe net purchases of assets acquired by investment funds. Since wefocus on actual portfolio choices of individual funds, rather than onbroad financial accounts sectors, our analysis is distinctly differentfrom that of Albertazzi, Becker, and Boucinha (2016) and Koijen etal. (2017). These studies provide an important wide-angle view ofportfolio rebalancing effects between broad sectors. Our more micro-focused analysis can account for heterogeneity in responses to assetpurchases across the sector and control for the extent of individualexposures to the policy.

By analyzing funds’ net purchases of assets, as distinct fromchanges in the value of their holdings, it is possible to identifyactive rebalancing behavior as distinct from passive rebalancing asdescribed in Bubeck, Habib, and Manganelli (2017) (this was notpossible, for example, in the analysis of Joyce et al. 2011). Sincefunds can be heavily mandated to hold certain types of assets, wealso explore whether some of the adjustment to the asset purchaseprogram is shared by fund investors. This is achieved through ananalysis of redemptions and issues. The detailed information aboutthe strategies of funds in our sample also helps us to disregard fundsthat do not have the flexibility to adjust in response to the program(e.g., exchange-traded funds and index trackers). The analysis of thisdetailed data should enable an identification of portfolio rebalancingeffects if they in fact exist.

While our analysis is entirely focused on the shadow bankingsector within one country, we cover a large fraction of all euro-area investment fund activity (i.e., roughly 17 percent of total assetsunder management in the region). This focus is warranted since thesector has grown significantly during the period covered by the assetpurchase program and it is a sector that is not as well understood—and not as well regulated—as the traditional banking sector. Thereis some evidence that banks have been an important channel forthe transmission of effects from asset purchases (e.g., Altavilla,Canova, and Ciccarelli 2016; Altavilla, Boucinha, and Peydro 2017;

facility), complex market expectations adjustments in response to signaling, andpolicy initiatives outside the EA (e.g., tightening in the United States).


and Carpinelli and Crosignani 2017), but we do not uncover stronglinks between our sample of investment funds and EA banks.

We employ a dynamic panel regression approach taking fulladvantage of the large cross-sectional sample of funds while control-ling for individual differences that are unrelated to the EAPP pro-gram. This methodology circumvents violations of classical regres-sion conditions required to deliver unbiased estimates of effects.We also control for the fundamental determinants of rebalancingassociated with fluctuations in macroeconomic variables.

Our results provide evidence of significant rebalancing towardsbonds issued by deposit-taking corporations but only after the paceof purchases was raised from €60–€80 billion per month. We findthat investors rebalance away from funds focused on holding theassets targeted by the purchase program (in the case of EA gov-ernment bonds, there is a move into shorter maturities—those noteligible for purchase under the EAPP). In terms of the maturitiesand currencies of newly purchased assets, we observe a move intonon-PSPP assets with longer maturities, away from assets denomi-nated in euros and into foreign-issued fixed-income assets. We findlittle evidence of a rebalancing towards equities.

It is important at the outset to acknowledge that heterogeneousviews exist about the likely success, adequacy, and legitimacy of theEAPP.5 Advocates for the program cite the likely tangible improve-ments in fiscal sustainability that would result from reductions interm and risk premiums. These would ultimately sow the seeds offuture growth and inflation. Opponents are of the view that thepurchase program (in conjunction with other ECB initiatives suchas lending operations and Outright Monetary Transactions) mightproduce excessive risk-taking and reduce incentives to make struc-tural reforms within the banking sector and in fiscal policy. This

5Firstly, Eggertsson and Woodford (2003) provide irrelevance propositions forcertain channels of such policies. Secondly, Ball et al. (2017) argue that the policyis too timidly applied and should be extended to the purchase of equities andcombined with a higher target for inflation. Thirdly, previous bond-buying poli-cies have been challenged in constitutional courts. The lawmaker Peter Gauweilerfrom the Christian Social Union party in Bavaria took a case to Germany’s Fed-eral Constitutional Court in 2013/14 concerning the legitimacy of the OutrightMonetary Transactions program.


may contribute to lower expected long-term growth and increasedfuture instability.

Assessing the merits of these opposing views is beyond the scopeof our analysis except in respect to how investor behavior can beused to infer a willingness to bet on the success of the program.Portfolio choices reflect changes in risk aversion, investor percep-tions of likely future risk exposures, and views about the relativefuture growth prospects of different economic sectors and economies.A valid inference from the revealed preference of investment funds(and investors) for the increased holding of non-euro-denominatedassets is that there is some reluctance to bet on the efficacy of theasset purchase program. This is of course a sample-specific result.

The paper is organized as follows. In the following two sectionswe discuss the modalities and operation of the purchase program sofar and consider related literature. This is followed by a detailed dis-cussion of the econometric specification we use to identify effects ofthe purchase program. We then consider the data and results. Thisis followed by a discussion of robustness analysis and conclusions.

2. The Extended Asset Purchase Programme So Far

The Eurosystem’s Extended Asset Purchase Programme wasannounced on January 22, 2015. The announcement was, to someextent, expected since normal policy tools (and some non-standardmeasures) had already reached the limits of their effectiveness. Therewere significant repayments of funding obtained in previous long-term refinancing operations (LTROs with three-year term) coincid-ing with the first of two new targeted long-term refinancing opera-tions (TLTROs), and it was feared in some quarters that TLTROswould be used as replacement funding rather than for new lending.6

Macroeconomic conditions had also remained subdued, and actualand expected inflation were well below target. The headline rateof inflation had remained stubbornly below 1 percent throughout

6Early repayment of funding obtained through three-year LTROs is discussedin box 3 of the February Monthly Bulletin of ECB (2013). Andrade et al. (2017)find that the three-year LTROs supported lending levels (they study the caseof French banks). The Fitch rating agency issued a statement in October 2014expressing a view that TLTROs would continue to ease funding conditions butwould be unlikely to affect lending.


2014, and it reached a low of 0.3 percent in November (0.7 percentif energy and food were excluded).

So conditions were ripe for a further extension of non-standardmeasures into a full-blown quantitative easing (QE) programthrough direct purchases of assets. The main component of this hasbeen the Public Sector Purchase Programme (PSPP). At the timeof the initial announcement, the Eurosystem made a commitment topurchase more than €1 trillion of securities over an eighteen-monthperiod. In December 2015 the termination date of the program wasextended from September 2016 to March 2017 and more assets wereincluded in the purchasable category (i.e., regional and local gov-ernment bonds). In March 2016 the size of the program was scaledup from €60–€80 billion per month. These changes were combinedwith two adjustments in the deposit facility rate (firstly, in Decem-ber 2015 to −30 basis points and then in March 2016 to –40 basispoints), which also broadened the universe of assets that could bebought.

The program was further extended to include investment-gradecorporate bonds in March 2016. By the end of 2016, the Eurosys-tem’s balance sheet contained €1,654 billion of assets for mone-tary policy purposes, and this included a €105 billion legacy fromthe Securities Markets Programme (specifically, government bondsissued by Greece, Ireland, Italy, Portugal, and Spain).7 By the endof April 2017, the ECB had purchased about €1.5 trillion under thePSPP component of the EAPP. Approximately €1.3 trillion of thisconsists of national government and agency bonds (see figure 1). TheABS and CBPP3 elements remain small, at roughly €24 and €216billion by the end of April 2017.

In terms of the modalities of the program, the PSPP is splitbetween government (central and local) and supranational in an 88percent to 12 percent ratio. The capital key is used to target inter-ventions across the issuer countries. This implies that countries withlarge GDP, large populations, and smaller national debt levels are

7The SMP was terminated on September 6, 2012 at the same time thattechnical features of Outright Monetary Transactions (OMT) measures wereannounced. SMP securities are held until maturity, and the outstanding stockfrom this program declined due to redemptions at a rate of about €20 billion peryear.


Figure 1. Eurosystem Holdings under the ExpandedAsset Purchase Programme

disproportionately affected by the purchases relative to outstand-ing debt. Bonds purchased must be BBB or better (i.e., at leastinvestment grade) with maturities between two and thirty years.Purchases cannot exceed 33 percent of the outstanding issued debtof a sovereign and cannot exceed 25 percent of a particular issue. Acondition applied to purchases, until its removal in December 2016,was that the yield to maturity of the purchased assets had to exceedthe ECB’s deposit facility rate at the time of purchase.

3. Literature

Much of the empirical research on the effects of asset purchasesby central banks has focused on the immediate (flow) and per-sistent (stock) price effects (see Gagnon et al. 2010, 2011; Chris-tensen and Rudebusch 2012; D’Amico et al. 2012; Joyce and Tonks2012; D’Amico and King 2013; Altavilla, Carboni, and Motto 2015;Andrade et al. 2016; Eser and Schwaab 2016; Haldane et al. 2016;and DeSantis and Holm-Hadulla 2017). The exchange rate effects areexamined by Glick and Leduc (2013). Volatility effects of QE havebeen analyzed by Bollerslev, Tauchen, and Zhou (2009), while equityprice effects and corporate bond spreads are examined in Benfordet al. (2009). The wider macroeconomic effects have been analyzed


by, for example, Baumeister and Benati (2013), Weale and Wieladek(2014), and Engen, Laubach, and Reifschneider (2015).

There are relatively few papers that have tried to get insights intothe portfolio rebalancing process by looking directly at the portfolioallocation decisions of financial intermediaries. Most portfolio rebal-ancing studies focus on flow-of-funds (financial accounts) sectors,including the investment fund category (e.g., Carpenter et al. 2013and Hogen and Saito 2014). Joyce, Liu, and Tonks (2015) extendsuch an analysis to explore the heterogeneous impact of QE on insur-ance companies and pension funds based on fund characteristics.Fratzscher, Lo Duca, and Straub (2016) conduct an analysis of theinternational spillover effects of EA unconventional monetary policy(UMP), up to the end of 2012, using daily investment fund flows.This is based on a fraction of flows based on funds sampled for theEPFR data set (Emerging Portfolio Fund Research). These authorsfind that UMPs led to a rise in global markets through increasedconfidence and lower risk premiums. Excluding the announcementof OMT, most UMPs led to some weakening of the euro. Overall theyfound little evidence of portfolio rebalancing effects through inter-national investment flows. Bubeck, Habib, and Manganelli (2017)also utilize the EPFR data source to assess portfolio rebalancingeffects in response to euro-area monetary policy actions and findmuted active responses but strong passive effects due to price andexchange rate outcomes.

Carpenter et al. (2015) find that sellers to the Federal Reservewere mainly households and other non-bank financial institutionscomprising hedge funds, broker-dealers, and insurance companies.Those who sold to the Federal Reserve subsequently purchased cor-porate bonds. Hogen and Saito (2014) find that Japanese banks andforeign investors were the largest sellers of government bonds to theBank of Japan and rebalanced towards bank loans, equity securi-ties, and corporate bonds. The micro-level analysis of Joyce, Liu,and Tonks (2015) indicates that sales of U.K. gilts to the Bank ofEngland by U.K. insurance companies and pension funds gave rise toreinvestment in corporate bonds. It is worth mentioning that Joyce,Liu, and Tonks (2015) do not possess the transactions data for theICPF entities and rely on the change in value of the holdings. Theycould therefore be identifying a valuation effect rather than a changein portfolio composition.


Another literature examines the impact of asset purchases onbank lending and profitability. Altavilla, Canova, and Ciccarelli(2016) uncover significant bank lending effects for the case of TLTROand APP measures in the long run, while Carpinelli and Crosignani(2017) find a positive effect of LTROs (those with three-year term)specifically for banks hit by funding dry-ups. The strength of sucheffects of course depends on the health and resilience of banks, theregulatory environment, and the economic conditions in the mar-kets into which banks traditionally lend (see Altavilla, Pagano, andSimonelli 2017; Bergant 2017; and Carpinelli and Crosignani 2017).An important point made by Valiante (2015) is that the low interestrate environment (and quantitative easing itself) may have producedchallenging conditions for bank profitability. This is challenged byAltavilla, Boucinha, and Peydro (2017), who consider a wider trade-off of effects and find that a decrease in short-term interest ratesdoes not necessarily imply lower bank profitability overall. However,there remains some concern, based on squeezed net interest marginsat low rates and at longer maturities, about the ability to engagein profitable and prudent lending. It is therefore valid to questionwhether banks alone can be relied upon to be the sole conduit ofunconventional monetary policy effects.

Two recent papers on portfolio rebalancing as a result of theECB’s Extended Asset Purchase Programme include Albertazzi,Becker, and Boucinha (2016) and Koijen et al. (2017). In Koijen et al.(2017), the effects of the EAPP have been analyzed for broad sectorsby using the holdings data of households and institutional investors.Their results so far only pertain to the period from 2013:Q4 to2015:Q4. They examine how rebalancing of portfolios affects riskand duration exposures.

They find that there is a strong home bias in holdings of PSPPassets by vulnerable countries for all sectors (this is similar tofindings by Hau and Lai 2016). Banks in vulnerable countries aretherefore disproportionately exposed to sovereign risk. The signif-icant results concerning responses to the EAPP purchases can bedescribed as follows: (i) foreign holders were most elastic in theirresponse to the program, (ii) the ECB buys 1.5 percent of durationrisk each month and reduces risk mismatch, (iii) there has been alarge reduction in debt issuance by banks in the euro area—suchbonds were held by foreign and vulnerable-country banks, and (iv)


the impact on EAPP asset prices has been 13 basis points but with alot of heterogeneity across country of issuance and maturities. Ouranalysis is not as wide ranging as theirs, but it concerns a moredetailed examination at a micro-level where cross-holdings are lesslikely to confound the results.

Albertazzi, Becker, and Boucinha (2016) examine the portfoliorebalancing of broad sectors (based on national financial accountssectors) using an identification based on the size of the valuationincrease experienced at the start of the EAPP. In the regressionexercise, the weight of specific securities in sectoral portfolios isinteracted with the price change of the security and this is, in turn,interacted with a dummy for the post-EAPP implementation period.They focus attention on newly issued debt securities. This is war-ranted because there must be equality in the rebalancing acrossinvestor groupings when the full population of investors is examined.Albertazzi, Becker, and Boucinha (2016) also consider the change inrisk (measured by ratings and residual maturity). They comparethe effects of EAPP across vulnerable and less vulnerable euro-areacountries.

While Albertazzi, Becker, and Boucinha (2016) find no statisti-cally significant relationship between portfolio rebalancing patternsacross sectors due to the extent of exposure to initial APP effectsfor the euro area as a whole, they do in fact find a relationship forsectors in “the more vulnerable countries” where more credit riskand increased exposure to non-financial corporation (NFC) debt isfound. They also find that banks that were most affected by APP(located in less vulnerable countries) increased their lending to NFCsand lowered the price of their credit to some sectors.

In our analysis, the question addressed is a bit different fromAlbertazzi, Becker, and Boucinha (2016). We examine how the fund-level portfolios within one sector (investment funds) domiciled in onecountry have adjusted when the supply of one asset class is reduced.There is a wide variety of assets that are not accounted for by theportfolios we examine (i.e., we don’t have the entire universe of assetsin our study and we do not have the entire population of investors).Since the value of many assets continually fluctuates, funds willrebalance portfolios to adjust their ex ante exposures. Also, sincevalues depend on the discounted value of returns—and these havequite variable fundamental determinants—it is possible to have large


valuation changes within a sector portfolio that is unrelated to thesupply of securities in circulation. In this context unanticipatedlosses on existing holdings will often be a primary determinantof future rebalancing behavior. Such confounding effects are, forexample, likely to pervade the analysis of Albertazzi, Becker, andBoucinha (2016). Our approach avoids much of this spurious effectby focusing on net purchases. We also control for the fundamen-tal determinants of valuation associated with fluctuations in macrovariables. These are now outlined as part of our modeling approach.

4. Model

Our aim is to examine how the ECB’s QE policy affected the within-sector rebalancing of investment funds (IFs). To this end, we focuson three different dimensions of portfolio composition, and we allowfor heterogeneities in the response to QE amongst different types offunds according to a broad set of fund characteristics.

We estimate the following panel regressions:

xit = αi + γxit−1

+ φ1QE antic + φ2QE first + φ3QE second

+ β1Debt nett + β2fin controljt

+ εit

(1)

lxit = αi + γxit−1

+ φ1QE antic + φ2QE first + φ3QE second

+ δ1QE antic ∗ Fund charactjit

+ δ2QE first ∗ Fund charactjit

+ δ3QE second ∗ Fund charactjit

+ β1Debt nett + β2fin controljt

+ β3Fund charactjit

+ εit,

(2)

where the dependent variable is the portfolio proportion of someasset class stated in percentage points for each investment fund


i and for quarter t. The assets are categorized into subsets alongthree dimensions of portfolio composition: (i) “Type and/or Issuer”of the asset; (ii) “Original Maturity” of the asset; and (iii) “Cur-rency of Denomination” of the asset at the time of issue. The firstof these broad categories is further subdivided into cash; equityissued by deposit-taking corporations; equity issued by other institu-tions; derivatives; security borrowing; overdrafts; other assets; bondsissued by governments; bonds issued by deposit-taking corporations;bonds issued by non-financial corporations; bonds issued by moneymarket funds, IFs, or financial vehicle corporations; and bonds issuedby other entities not already mentioned. The “Original Maturity”category is subdivided into assets with term to maturity less thanone year, assets with term to maturity between one and two years,assets with term to maturity greater than two years, and assets withno planned term to maturity. For the final broad category, “Currencyof Denomination,” we subdivide into the three main currencies:USD, euros, and GBP.

To obtain a comprehensive overview of the dynamics of portfoliocomposition, we examine both the end-quarter stock positions andthe within-quarter flows (i.e., transactions/net purchases expressedas a percentage of total assets) for each asset class. The stock posi-tion is subject to revaluation effects, and these tend to dominatethe rebalancing effects. Purchases and sales of assets are recorded atmarket value at the time of the transactions so we can rely on thedifference between gross buy and sell transactions to reflect portfoliorebalancing net of revaluations.8

In the first regression, we relate the portfolio rebalancing vari-able to QE dummy variables, namely, QE antic, QE first, andQE second. These dummy variables select the following time peri-ods, respectively, for which QE effects (or the effects of its anticipa-tion) are estimated: 2014:Q4, from 2015:Q1 to 2016:Q1, and from

8In particular, the stock is the share of asset class x that the fund i investsat the end of period t over total assets of fund i at the end of period t. The flowis the share of net transactions for asset class x at the end of period t for fund iover total assets at the beginning of period t for fund i. Net transactions are themarket value of purchases and sales of a security on the dates of each transaction.A purchase implies an increase in the position, and sales imply a decrease in theposition. Short-selling a security is a decrease in position (i.e., sale).


2016:Q2 to 2016:Q3, respectively.9 Following the work by Carpen-ter et al. (2013) and Joyce, Liu, and Tonks (2015) and to controlfor non-QE macroeconomic contributors to portfolio rebalancing aswell as supply-side effects of new issuances, we include the follow-ing explanatory variables in the regressions: (i) the net issuance ofdebt in the euro area in billions (Debt net) and (ii) the followinglist of macroeconomic controls (collectively denoted fin controljt):the U.S. ten-year Treasury yield adjusted to constant maturity(US long yield), the ten-year corporate spread (us corp spread2),and the VIX (us v ix vol).

In the second regression, we add the interaction of QE vari-ables with specific fund characteristics (Fund charactit), which areinvestor withdrawals relative to beginning-of-period net asset value(NAV) (redeem nav) and investment inflows relative to beginning-of-period NAV (issuance nav), the fund leverage as a proportionof NAV (leverage nav), and the average of sales and purchases ofsecurities relative to NAV (turnover nav).10

We retain for analysis only those funds that are actively man-aged, and we exclude those funds that have a daily turnover relativeto net asset value in excess of 17 percent. This is to avoid the inclu-sion of funds that engage in high-frequency trading.11 In order toease the interpretation of the results, all the variables are centeredaround the pre-QE period average (2014:Q1–2014:Q3).12

9Recall, the combined monthly purchases of private and public securities con-sisted of €60 billion from March 2015 up until March 2016. Starting in April 2016,the monthly purchase increased to €80 billion on average.

10We also try to interact the following fund characteristics which, however,were not significant: size in €billions (based on NAV) (size), closing position ofasset denominated in currency other than reporting currency as a share of totalassets (closing position) (foreigncurr ta), sum of Financial Services Fees Euro +Other Professional Fees Euro + Other Operating Expenses Euro)/Total assets(expense ta), sum of all increases and decreases of derivatives transactions. Bothassets and liabilities are included (derivative nav).

11The value associated with the highest decile of daily turnover relative to NAVis 17 percent. A total of 529 observations involving 143 funds are dropped. Thiscutoff occurs at the point where the positive skew of the distribution is maximallyreduced. We also deleted the upper and lower 0.5 percentile of all the variablesto exclude influential outliers.

12This ensures that the dummy variable coefficients can be interpreted with-out subtraction from an intercept term. If this adjustment is not made, the Stata


Given that our dependent variables are serially correlated, weinclude a lagged dependent variable in our regression, and this leadsnaturally to estimation of a dynamic panel. We employ the dynamicpanel methods pioneered by Holtz-Eakin, Newey, and Rosen (1988)and further developed by Arellano and Bond (1991), Arellano andBover (1995), Blundell and Bond (1998), Baum, Schaffer, and Still-man (2003), Windmeijer (2005), and Bai and Ng (2010). Specifically,we use an Arellano-Bond/Arellano-Bover dynamic panel which is adifference generalized method of moments (DGMM) where the indi-vidual fund fixed effect is purged using orthogonal deviation andthe lagged level variables are used to instrument the transformedequation.13 All individual fund characteristics and the lagged val-ues of the dependent variable are considered endogenous and areinstrumented with their second and third lags. QE variables and theforeign financial variables are considered strictly exogenous.14

The above model is applied to two different subsamples of invest-ment funds based on whether or not they have held a significantfraction of their portfolio in the form of PSPP assets. Specifically,we denote funds as PSPP holders if they have, in any quarter of oursample, held at least 10 percent of their portfolio in PSPP assetsbought by the ECB between March 2015 and June 2016. The sec-ond sample (non-PSPP holders ) includes funds that have never heldPSPP assets.15 In our discussion of results we focus almost entirelyon the case of PSPP holders, as they are the only potential sellersto the ECB. We also rely more on the results explaining net pur-chases rather than the proportional holdings. We do not generally

procedure includes an intercept, and this is awkward to interpret since it implic-itly applies to all periods.

13As compared to Arellano and Bond (1991), the transformed equation in Arel-lano and Bover (1995) is obtained by subtracting the average of all availablefuture observations instead of differencing. The model is particularly suitable forunbalanced panel, as it has the virtue of preserving the sample size.

14For the QE variables we also use the pass-through option in order to avoidtheir transformation and to guarantee that they are instrumented with theirown value. For all regressions we use a robust two-step estimator where thestandard covariance matrix is robust to panel-specific autocorrelation and het-eroskedasticity and the standard error is correct using the Windmeijer finitesample methodology.

15The samples are identified using ISIN-level intervention data provided by theFinancial Market Division of the Central Bank of Ireland. The use of those dataguarantees no uncertainty in the identification of PSPP holders.


find significant results for the case of proportional holdings (i.e., pas-sive effects) of PSPP holders or for the rebalancing of the non-PSPPholders.16

Summary statistics for the dependent variables and the fundcharacteristics (including the “within” and “between” standard devi-ations) for the PSPP holders are provided in tables 1, 2, and 3. Theresults reveal that most of the variance is explained by cross-unitvariation, while the within variation is limited. The only exceptionis for redemption and issuance, which display a substantial withinvariation.

Before discussing regression results, it is worth noting fromdescriptive statistics in table 3 that there is significant correlationbetween the proposed explanatory variables in equation (2). Thepooled correlations shown in the top panel of table 3 reveal thatturnover is highly correlated with leverage (0.63), foreign currencyposition (0.26), and issuance (0.21). These raw correlations are dif-ficult to interpret—specifically, we do not know whether they reflectcommon time or cross-sectional variation.

5. Data and Results

5.1 Data

Our analysis focuses on the large investment fund industry domi-ciled and reporting in Ireland.17 The Central Bank of Ireland’sStatistics Division collects quarterly balance sheet information andmonthly investment fund information through investment fundreturns. These data provide a comprehensive overview of all funds’quarterly accounts characteristics, gross buy-and-sell transactions,and positions vis-a-vis residents and non-residents by reporting cur-rency. Accounting information includes—amongst other items—thesecurity-by-security information on holdings of equities, debt securi-ties, and derivatives; profits and losses per period on an accrualsbasis; interest, dividends, rents, and other income. Our data set

16In the non-PSPP sample we also include dummy variables that capture thetype of fund (bond, equity, hedge, money market fund, mixed, other, or realestate).

17The total asset value of the sector in 2016:Q3 is about €2 trillion.


Tab

le1.

Des

crip

tive

Sta

tist

ics

ofP

SP

PH

older

s/D

epen

den

tV

aria

ble

s:Sto

ck/M

ean

2014

:Q1–

2016

:Q3

Por

tfol

ioCom

posi

tion

byA

sset

Typ

e—M

ean

and

Sta

nda

rdD

evia

tion

s

cash

equity

dtc

equity

othe

rsde

riva

tive

sse

cbor

rov

erdr

otha

ssbo

nd

gov

bond

dtc

bond

nfc

bond

ofibo

nd

othe

rs

Mea

n0.

280.

0510

.06

4.08

0.06

3.61

3.02

59.0

75.

494.

070.

027.

88σ

o1.

561.

8418

.59

11.1

71.

016.

547.

0331

.55

8.79

7.59

.24

11.2

8σ

b1.

581.

8118

.09

10.7

90.

776.

835.

5830

.16

8.96

7.16

.30

10.7

7σ

w0.

950.

634.

473.

94.5

42.

904.

119.

992.

282.

710.

163.

31

Por

tfol

ioCom

posi

tion

byO

rigi

nal

Mat

uri

tyan

dCurr

ency

—M

ean

and

Sta

nda

rdD

evia

tion

s

less

1yea

rle

ss2y

ear

mor

e2ye

arpe

rpcu

rrusd

curr

euro

curr

gbp

Mea

n7.

911.

6672

.35

0.02

18.1

668

.24

4.95

σo

12.7

64.

3627

.94

0.14

22.5

434

.38

0.10

2σ

b11

.80

3.80

26.7

80.

1322

.18

34.1

69.

61σ

w6.

622.

4310

.08

0.08

5.39

4.65

2.56

Note

:T

his

table

report

sth

edes

crip

tive

stati

stic

sfo

rth

edep

enden

tva

riable

suse

din

equati

on

(1)

for

quart

erly

obse

rvati

ons

from

2014:Q

1to

2016:Q

3(b

efore

dem

eanin

g).

The

standard

dev

iati

on

stati

stic

s,σ

o,σ

b,and

σw

,re

fer

toov

erall,bet

wee

n,and

wit

hin

vari

ati

on,re

spec

tivel

y.


Tab

le2.

Des

crip

tive

Sta

tist

ics

ofP

SP

PH

older

s/D

epen

den

tV

aria

ble

s:Flo

ws/

Mea

n20

14:Q

1–20

16:Q

3

Por

tfol

ioCom

posi

tion

byA

sset

Typ

e—M

ean

and

Sta

nda

rdD

evia

tion

s

cash

equity

dtc

equity

othe

rsde

riva

tive

sse

cbor

rov

erdr

otha

ssbo

nd

gov

bond

dtc

bond

nfc

bond

ofibo

nd

othe

rs

Mea

n−

0.01

0.00

0.10

0.85

−0.

010.

31−

0.05

−1.

110.

200.

130.

000.

31σ

o0.

770.

494.

115.

610.

263.

956.

6211

.09

2.40

2.23

0.11

3.49

σb

0.63

0.28

2.29

4.59

0.14

1.80

3.73

8.56

1.35

1.31

0.07

1.69

σw

0.69

0.46

3.65

3.57

0.22

3.72

6.23

8.65

2.15

1.95

0.10

3.19

Por

tfol

ioCom

posi

tion

byO

rigi

nal

Mat

uri

tyan

dCurr

ency

—M

ean

and

Sta

nda

rdD

evia

tion

s

less

1yea

rle

ss2y

ear

mor

e2ye

arpe

rpcu

rrusd

curr

euro

curr

gbp

Mea

n0.

490.

04−

0.67

0.00

1.03

−1.

300.

31σ

o8.

651.

9612

.84

0.01

10.2

213

.36

4.12

σb

4.89

0.67

9.82

0.01

8.55

8.96

2.45

σw

8.02

1.88

10.1

80.

018.

5010

.69

3.69

Note

:T

his

table

report

sth

edes

crip

tive

stati

stic

sfo

rth

edep

enden

tva

riable

suse

din

equati

on

(1)

for

quart

erly

obse

rvati

ons

from

2014:Q

1to

2016:Q

3(b

efore

dem

eanin

g).


Tab

le3.

Des

crip

tive

Sta

tist

ics

and

Cor

rela

tion

sof

PSP

PH

older

s/Fund

Char

acte

rist

ics

size

fore

igncu

rrta

expe

nse

tale

vera

genav

turn

over

nav

deri

vative

nav

rede

mnav

issu

ance

nav

Var

iabl

eCor

rela

tion

s

size

1fo

reig

ncu

rrta

0.12

4∗∗

∗1

expe

nse

ta−

0.17

0∗∗

∗0.

0649

∗∗

1le

vera

genav

0.14

0∗∗

∗−

0.01

28−

0.14

1∗∗

∗1

turn

over

nav

0.14

7∗∗

∗0.

268∗

∗∗

0.12

1∗∗

∗0.

144∗

∗∗

1de

riva

tive

nav

0.18

0∗∗

∗0.

246∗

∗∗

−0.

0065

00.

276∗

∗∗

0.63

2∗∗

∗1

rede

mnav

−0.

0141

−0.

0729

∗∗

0.10

4∗∗

∗0.

0461

∗0.

176∗

∗∗

0.05

36∗

1is

suan

cenav

0.04

310.

0558

∗0.

0235

0.00

148

0.21

8∗∗

∗0.

0408

0.13

1∗∗

∗1

Des

crip

tive

Sta

tist

ics:

Mea

nan

dSta

nda

rdD

evia

tion

s:M

ean

2014

:Q1–

2016

:Q3

Mea

n26

.09

25.2

60.

4225

.50

2.31

32.1

42.

832.

81σ

o40

.30

28.7

20.

6181

.57

3.41

96.5

63.

826.

44σ

b36

.93

28.2

20.

6680

.06

3.33

107.

232.

855.

26σ

w9.

075.

730.

3032

.64

1.67

43.7

22.

815.

18

Des

crip

tive

Sta

tist

ics:

Mea

nan

dSta

nda

rdD

evia

tion

s:M

ean

2014

:Q1–

2014

:Q3

Mea

n26

.02

26.7

20.

4517

.94

2.14

21.9

32.

343.

17σ

o36

.97

29.6

00.

5545

.60

3.26

61.1

02.

765.

49σ

b36

.31

29.7

10.

6243

.50

3.51

66.4

22.

975.

53

Note

s:T

he

firs

ttw

opart

sof

the

table

report

corr

elati

on

and

the

des

crip

tive

stati

stic

sfo

rth

ein

ves

tmen

tfu

nd

chara

cter

isti

csfo

rth

equart

erly

obse

r-va

tions

from

2014:Q

1to

2016:Q

3(b

efore

dem

eanin

g).

The

standard

dev

iati

on

stati

stic

sσ

o,

σb,and

σw

,re

fer

toov

erall,bet

wee

n,and

wit

hin

vari

ati

on,

resp

ecti

vel

y.T

he

last

part

ofth

eta

ble

report

sth

edes

crip

tive

stati

stic

sfo

rth

equart

erly

obse

rvati

ons

from

2014:Q

1to

2014:Q

3.In

this

case

,th

est

andard

dev

iati

on

stati

stic

s,σ

oand

σb,re

fer

only

toth

eov

erall

and

bet

wee

nva

riati

on.


spans the quarters from 2014:Q1 to 2016:Q3 and includes all invest-ment funds categorized as equity, bond, mixed, hedge, real estate,money market, and others. As we mentioned previously, we struc-ture the analysis by two subsamples depending on whether fundshave a substantial holding of PSPP assets. We obtained all event-by-event PSPP intervention data by ISIN identifiers from the MarketsDivision of the Irish Central Bank.

Table 4 provides a first overview of the two samples. This revealsthat PSPP asset holders are mainly dominated by bond and mixedfunds, while the non-PSPP asset holders are dominated by equityfunds.18 Figure 2 provides a deeper look into the portfolio of PSPPholders and suggests that most of the funds hold a very small per-centage of PSPP assets (right axis). Also, combining the share ofPSPP holdings with the interquartile range of their value (left axis),the graph suggests that large PSPP holders are also large funds.19

Figure 3 plots the average end-of-quarter proportional portfoliocompositions of the PSPP asset-holding funds according to assettype/issuer, maturity, and currency (as defined above). In the caseof the holdings by asset type, the graphs include only the main types(i.e., cases where, on average, the asset type represents more than4 percent of the total assets held by the funds).20 We see that, onaverage, portfolios of PSPP asset-holding funds are heavily concen-trated on government bonds denominated in euros with maturitylonger than two years. The time profile of proportional holdingsexhibits limited movement. In fact, the proportional holdings of gov-ernment bonds has clearly increased since the start of the ECB’sasset program. This suggests that funds have persistent preferences,but it does not necessarily reflect the true impact of the EAPP.The increase in the share of assets allocated to government bondscould merely be the consequence of revaluation effects combinedwith redemptions.

18The total asset value of the PSPP holders in 2016:Q3 is about €340 billion,while for non-PSPP holders it is about €1.6 trillion.

19For example, size can vary a lot—from about 80 million to 1 billion for thebar between 0.6 and 0.7—which are funds that held between 60 and 70 percentof their portfolio in PSPP assets.

20Figures are the values before demeaning and after removing funds withturnover relative to NAV > 17 percent.


Tab

le4.

Num

ber

ofFunds

by

Inve

stm

ent

Str

ateg

y

2014:Q

12014:Q

22014:Q

32014:Q

42015:Q

12015:Q

22015:Q

32015:Q

42016:Q

12016:Q

22016:Q

3Tota

l

PSPP

Hol

ders

Bon

d11

411

311

911

712

112

312

913

213

414

114

11,

384

Equ

ity

00

00

00

00

11

13

Hed

ge1

11

23

22

34

55

29M

ixed

4447

4747

4149

5150

4852

5352

9O

ther

56

67

88

88

911

1187

Tot

al16

416

717

317

317

318

219

019

319

621

021

12,

032

Non

-PSPP

Hol

ders

Bon

d37

736

738

337

638

539

240

641

240

641

642

74,

347

Equ

ity

1,08

61,

070

1,08

11,

083

1,09

51,

119

1,14

21,

164

1,18

31,

188

1,22

212

,433

Hed

ge52

750

149

944

544

944

946

244

945

244

542

95,

107

MM

F26

1825

1923

2119

1918

2021

229

Mix

ed39

841

741

842

542

443

246

446

947

849

349

24,

910

Oth

er31

631

631

933

235

636

537

938

339

540

840

73,

976

Rea

lE

stat

e43

5663

7787

8999

9311

712

413

698

4Tot

al2,

773

2,74

52,

788

2,75

72,

819

2,86

72,

971

2,98

93,

049

3,09

43,

134

31,9

86

Sourc

e:A

uth

ors

’ca

lcula

tions.

Note

:T

his

table

report

sth

enum

ber

offu

nds

inth

etw

osa

mple

sby

inves

tmen

tst

rate

gy.


Figure 2. Distribution of Share of PSPP Assets Held byEach Fund in Percent (right axis) and Interquartile Rangeof the Value of the PSPP Assets in the Portfolio of Each

Fund (left axis)

Figure 3. PSPP Holders: Portfolio Composition/Stock

A. Asset Type B. Maturity

C. Currency


Figure 4. PSPP Holders: PortfolioComposition/Net Purchases


C. Currency

More reliable insights about rebalancing can be added from anexamination of transactions for the same asset classes. Figure 4shows net purchases, and this produces quite a different picture fromthat of the proportional holdings. On a net basis, funds have clearlysold euro-denominated government bonds (particularly those withmaturity longer than two years). They also seem to have movedtowards other types of bonds and to derivatives.21 This graphicalanalysis seems to support the presence of a portfolio rebalancingchannel. However, it is important to assess whether this behaviorcould be due to normal rebalancing prompted by non-QE-relatedmacroeconomic developments. We now consider the results of a

21Results for non-PSPP holders are reported in figure 5 and figure 6. Over-all, the non-PSPP funds seem to hold mainly equities denominated in U.S. dol-lars. Their portfolios are spread across all maturities. Also, net purchases do notsuggest a clear rebalancing pattern.


Figure 5. Non-PSPP Holders: PortfolioComposition/Stock


C. Currency

Figure 6. Non-PSPP Holders: PortfolioComposition/Net Purchases


C. Currency


panel regression analysis that quantifies the rebalancing sensitivityto asset purchases for the different phases of the program while con-trolling for fund characteristics, macroeconomic developments, andfund fixed effects.

5.2 Results

Table 5 provides results for the dynamic panel regression in equa-tion (1) explaining net purchases of the six largest asset classes inthe portfolios of funds defined as significant holders of PSPP assets.In this first regression we exclude fund characteristics in the controlset. We use the Arellano-Bover forward orthogonalization approachto purge fixed effects. In general, the results have intuitive appealand their interpretation coincides with our expectations. Overall,there is only significant evidence of rebalancing during the secondphase of the purchase program and only two assets in which sig-nificant QE rebalancing effects can be identified. The coefficientson the second-phase QE dummies (QE second) indicate that netpurchases of government bonds are more negative than usual, andnet purchases of bonds issued by deposit-taking corporations aregreater than usual, when the program was scaled up. The signifi-cant negative flows out of government bonds in the second phase ofthe program probably reflects both the increased intensity of pur-chases and an increased willingness to divest of APP assets at thefirst sign of improving economic conditions (which could signal anearlier timing of tapering than originally expected). We find no sig-nificant rebalancing effects during the anticipation period and nosignificant rebalancing flows in any QE period towards equities orderivatives. It is worth noting that equities and derivatives repre-sent only a very small proportion of the portfolio of the funds (seetable 1).

While this first regression provides limited but broadly sup-portive evidence for rebalancing, it should be noted that, despitethe presence of a lagged dependent variable, the regression residualremains significantly autocorrelated in two cases (there is clear evi-dence of AR(2) in the differenced error, implying AR(1) in levelsfor the case of bonds issued by governments and by deposit-takinginstitutions). Furthermore, the tests for validity of the GMM instru-ments indicate weakly specified models in the cases of bonds issued

Vol. 15 No. 5 The Portfolio Rebalancing Effects 25Tab

le5.

PSP

PSam

ple

:N

etP

urc

has

es

Var

iable

sEqu

ity

othe

rsD

eriv

ativ

esBon

dgo

vBon

ddt

cBon

dnfc

Bon

dot

hers

QE

antic

0.19

50.

0108

−1.

079

0.03

030.

189

−0.

375

(0.3

17)

(0.1

91)

(1.0

90)

(0.2

16)

(0.1

58)

(0.3

11)

QE

first

−0.

106

0.23

4−

1.16

10.

173

0.23

4−

0.45

7(0

.284

)(0

.301

)(1

.510

)(0

.241

)(0

.225

)(0

.415

)Q

Ese

cond

−0.

697

0.10

8−

4.92

2∗∗0.

772∗∗

0.45

4−

0.18

0(0

.460

)(0

.330

)(2

.033

)(0

.355

)(0

.301

)(0

.573

)D

ebtne

t−

0.00

0860

−0.

0004

89−

0.01

30∗∗

∗0.

0006

340.

0007

410.

0011

3(0

.001

13)

(0.0

0080

5)(0

.003

87)

(0.0

0062

9)(0

.000

530)

(0.0

0089

9)U

Slo

ngyi

eld

−0.

439

−0.

0952

−3.

817∗∗

0.86

1∗∗0.

463

0.27

4(0

.434

)(0

.358

)(1

.795

)(0

.335

)(0

.319

)(0

.584

)us

corp

spre

ad2

−0.

203

−0.

0288

0.13

40.

171

0.01

040.

0480

(0.1

25)

(0.1

20)

(0.6

08)

(0.1

08)

(0.0

722)

(0.1

03)

usvi

xvo

l−

0.05

84∗

−0.

0351

∗−

0.14

60.

0317

0.00

343

0.00

891

(0.0

353)

(0.0

196)

(0.1

49)

(0.0

204)

(0.0

139)

(0.0

255)

L.d

epot

hers

−0.

159∗∗

−0.

266∗∗

∗−

0.02

06−

0.07

69−

0.11

2∗∗−

0.09

31(0

.076

4)(0

.060

8)(0

.065

2)(0

.060

5)(0

.049

6)(0

.075

0)

Obs

erva

tion

s1,

340

1,35

21,

339

1,34

01,

346

1,34

4N

umbe

rof

frfu

ndke

y17

417

417

417

417

417

4A

R1

p-va

lue

0.00

687

0.01

941.

25e-

050.

0008

600.

0005

540.

0020

1A

R2

p-va

lue

0.98

40.

724

0.09

290.

131

0.05

770.

961

Han

sen-

Jp-

valu

e0.

461

0.11

60.

0213

0.11

90.

380

0.45

7W

ald

p-va

lue

(Joi

nQ

E)

0.13

930.

6964

0.00

280.

0215

0.22

910.

1423

Note

s:T

his

tabl

ere

por

tspa

nel

regr

essi

onre

sult

sfo

rth

em

odel

desc

ribed

ineq

uati

on(1

)fo

rne

tpu

rcha

ses.

We

repor

ton

lyth

ere

sult

sfo

ras

set

clas

ses

for

whi

chth

epr

opor

tion

over

tota

las

sets

ison

aver

age

abov

e4

per

cent

.W

esu

btra

ctfr

omth

eva

riab

les

the

mea

nof

the

per

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.D

ebtne

tis

inbi

llion

sof

euro

s.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


by governments and by NFCs. These issues are ameliorated to asignificant extent in the more general specification now discussed.

Table 6 provides regression results for equation (2) where weinclude fund issuance, redemption, leverage, and turnover charac-teristics (both on their own and interacted with QE dummies)to explicitly allow heterogeneity in portfolio rebalancing behavioracross funds. In comparison with the first set of regression results, weobserve a similarly significant (but now larger) increase in the pur-chase of the bonds issued by deposit-taking institutions regardlessof the fund characteristics (i.e., a significant positive coefficient onthe QE second dummy in the Bonddtc column). The net purchase ofbonds of deposit-taking institutions is also large and statistically sig-nificant for the second QE period among the funds that are relativelymore engaged in the issuance of new fund units (i.e., a significantpositive coefficient on the Second iss dummy).

The significant negative effect found in our first regression in thecase of net purchases of government bonds on the non-interactedQE second dummy becomes insignificant in the more general spec-ification. However, we now see that the coefficient on redemptionon its own, redem nav , is negative and significant and there aretwo significantly negative coefficients on the interactive QE dum-mies (Firstrdm and Secondrdm), indicating net sales of governmentbonds by funds that experienced relatively more redemptions of fundunits generally and specifically during the QE-active periods. Thisis consistent with a coincidence in the rebalancing behavior of fundinvestors away from funds focused on holding government bondsand rebalancing away from government bonds by the funds them-selves. The coefficient on issuance on its own is significantly positive,but there is no significance for the interaction of issuance with QEdummies.

There is significant support for the rebalancing channel dueto movement into bonds issued by those classified as “other”(Bondother). Here we observe significantly more net purchases dur-ing each of the QE-active periods by funds experiencing relativelyhigher redemptions as well as those experiencing relatively moreissuance of their own units (i.e., significance of coefficients onFirstrdm, Secondrdm, Firstiss, and Secondiss). It is noteworthythat there was little significant evidence of rebalancing towards equi-ties (except for the case of funds with relatively more leverage) or


le6.

PSP

PSam

ple

:N

etP

urc

has

es/I

nte

ract

ion

with

Fund

Char

acte

rist

ics

Var

iable

sEqu

ity

othe

rsD

eriv

ativ

esBon

dgo

vBon

ddt

cBon

dnfc

Bon

dot

hers

QE

antic

0.29

9−

0.85

70.

845

0.15

90.

328

0.29

6(0

.612

)(1

.023

)(0

.992

)(0

.263

)(0

.455

)(0

.389

)Q

Efir

st−

0.15

90.

502

1.55

00.

550

0.18

10.

174

(0.6

12)

(0.5

83)

(1.2

78)

(0.3

44)

(0.3

11)

(0.5

64)

QE

seco

nd−

0.20

40.

0646

−0.

754

1.60

8∗∗

0.28

10.

616

(1.0

49)

(1.0

70)

(1.7

39)

(0.6

83)

(0.4

58)

(0.8

53)

Ant

rdm

−35

.73

76.9

8−

70.5

38.

807

−54

.19

39.1

4(3

2.57

)(5

4.50

)(5

6.58

)(3

4.73

)(4

0.61

)(3

3.01

)Fir

strd

m−

4.86

8−

20.6

4−

55.6

8∗−

0.79

0−

31.0

0∗∗∗

51.7

1∗∗∗

(22.

65)

(39.

11)

(32.

83)

(10.

76)

(10.

30)

(17.

06)

Seco

ndrd

m−

0.41

9−

28.6

9−

70.8

7∗∗

3.81

7−

22.2

1∗∗

57.3

0∗∗∗

(23.

03)

(36.

14)

(35.

29)

(8.5

30)

(9.6

80)

(17.

10)

Ant

iss

−1.

075

6.00

029

.54

0.42

2−

6.57

519

.20

(8.9

46)

(20.

51)

(24.

80)

(8.0

30)

(9.6

17)

(13.

34)

Fir

stis

s3.

069

3.42

215

.88

5.37

2−

9.33

718

.71∗

∗∗

(7.7

58)

(14.

21)

(21.

29)

(3.5

34)

(6.4

40)

(6.1

61)

Seco

ndis

s5.

620

1.18

56.

323

8.73

5∗∗

−10

.07

21.2

2∗∗∗

(8.7

83)

(12.

03)

(20.

44)

(4.1

95)

(7.2

56)

(7.2

02)

Red

emna

v−

12.0

424

.52

−97

.66∗

∗−

23.9

5∗∗

11.3

2−

63.5

9∗∗∗

(23.

85)

(27.

01)

(39.

69)

(11.

59)

(10.

15)

(15.

22)

Issu

ance

nav

20.2

114

.97

157.

5∗∗∗

22.2

4∗30

.86∗

∗43

.50∗

∗

(16.

03)

(21.

49)

(41.

87)

(11.

37)

(13.

69)

(18.

12)

L.d

ep−

0.20

6∗∗∗

−0.

261∗

∗∗−

0.01

78−

0.10

1∗−

0.10

6∗∗

−0.

0885

(0.0

733)

(0.0

684)

(0.0

403)

(0.0

526)

(0.0

512)

(0.0

865)

Obs

erva

tion

s1,

308

1,31

91,

307

1,30

71,

313

1,31

1N

umber

offr

fund

key

173

173

173

172

173

173

AR

1p-

valu

e0.

0072

10.

0065

90.

0001

963.

74e-

050.

0005

910.

0073

0A

R2

p-va

lue

0.97

80.

939

0.68

60.

0813

0.11

00.

538

Han

sen-

Jp-

valu

e0.

612

0.00

730

0.11

60.

189

0.65

50.

328

Wal

dp-

valu

e(J

oin

QE

)0.

9574

0.30

540.

0100

0.04

330.

8233

0.48

42

Note

s:T

his

table

repor

tspan

elre

gres

sion

resu

lts

for

the

mod

eldes

crib

edin

equat

ion

(2)

for

net

purc

has

es.W

ere

por

ton

lyth

ere

sult

sfo

ras

set

clas

ses

for

whic

hth

epro

por

tion

over

tota

las

sets

ison

aver

age

abov

e4

per

cent

.W

esu

btr

act

from

the

vari

able

sth

em

ean

ofth

eper

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.T

he

regr

essi

ons

incl

ude

also

deb

tnet

,fin

cont

rols

,tu

rnov

ernav

,le

vera

genav

,an

dth

eir

inte

ract

ion

wit

hth

eQ

Edum

mie

s.Sta

ndar

der

rors

are

inpar

enth

eses

.**

*,**

,an

d*

den

ote

p<

0.01

,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


derivatives, even though these assets are significant (if small) ele-ments of the PSPP holders’ portfolios. Overall, the results for thismore flexible regression specification are relatively supportive of arebalancing channel at work. However, the issuance currency of theassets substituted for PSPP assets in portfolios has yet to be consid-ered, so the exact transmission mechanism of the purchases is stillnot entirely clear (we return to this below).

Contrary to our prior expectations, there is evidence of a with-drawal from bonds issued by non-financial corporations (Bondnfc)in both QE-active periods by funds with relatively high redemptions.Thus, unlike the findings of studies of rebalancing behavior duringU.S., U.K., and Japanese QE programs, NFCs do not seem to ben-efit from portfolio rebalancing in the euro-area case (investment isinstead directed towards bank bonds and bonds of issuers classifiedas “other”).

The results just discussed hint at a rebalancing channel throughredemptions and issues in combination with portfolio rebalancingat the fund level. We also found some correspondence in the levelof leverage and QE effects. To assess these indirect channels, weexamine how redemptions, issues, leverage, and turnover for PSPP-holding funds was affected by EAPP. Table 7 provides results forregressions with a similar structure to that of equations (1) butwith redemption, issuance, leverage, and turnover (all relative toNAV) as the dependent variables. We see that redemptions rosemore significantly than issuances for the PSPP-holding funds, andthese effects are rising in magnitude over the course of the asset pur-chase program and its anticipation. The use of leverage seems to havebeen unaffected by the EAPP, and turnover only rose significantlyin the second QE-active period. Overall, the redemption, issuance,and activity evidence is consistent with a flight of investment awayfrom PSPP-holding funds.

The non-PSPP holders were not directly affected by the rela-tively high level of purchases of PSPP assets by the ECB because,by definition, they were not holding PSPP assets. However, the stan-dard rebalancing channel is expected to act indirectly through thehigher pricing (and lower yield) of assets that are close substitutesfor the assets purchased. To assess whether there is any evidencefor these indirect effects, we run the same regressions as above forvarious fund categories within the sample of funds that never held


le7.

PSP

PSam

ple

:R

edem

ption

,Is

suan

ce,Lev

erag

e,an

dA

ctiv

ity

Var

iable

sre

dem

nav

issu

ance

nav

leve

rage

nav

turn

over

nav

QE

antic

0.89

6∗∗

0.25

0−

0.59

70.

0376

(0.3

84)

(0.2

89)

(1.1

16)

(0.1

22)

QE

first

1.57

1∗∗∗

0.74

3∗∗

1.89

40.

0818

(0.3

94)

(0.3

17)

(1.2

86)

(0.1

62)

QE

seco

nd3.

044∗

∗∗1.

196∗

∗∗0.

400

0.66

5∗∗∗

(0.5

99)

(0.4

15)

(2.2

12)

(0.2

52)

Deb

tne

t0.

0052

7∗∗∗

0.00

138

−0.

0060

90.

0025

9∗∗∗

(0.0

0104

)(0

.000

913)

(0.0

0915

)(0

.000

518)

US

long

yiel

d2.

082∗

∗∗1.

348∗

∗∗1.

051

0.42

6(0

.562

)(0

.408

)(1

.893

)(0

.264

)us

corp

spre

ad2

0.56

4∗∗∗

0.42

8∗∗∗

0.29

70.

150∗

∗

(0.1

65)

(0.1

28)

(0.3

76)

(0.0

704)

usvi

xvo

l0.

0852

∗∗0.

0376

0.00

404

0.04

32∗∗

∗

(0.0

347)

(0.0

261)

(0.1

50)

(0.0

157)

L.d

ep−

0.06

80−

0.08

52∗

0.62

1∗∗∗

−0.

143

(0.0

574)

(0.0

461)

(0.0

443)

(0.0

887)

Obs

erva

tion

s1,

351

1,34

31,

334

1,32

8N

umbe

rof

frfu

ndke

y17

417

317

417

4A

R1

p-va

lue

0.00

0236

0.00

0134

0.01

010.

0156

AR

2p-

valu

e0.

290

0.19

80.

0600

0.82

2H

anse

n-J

p-va

lue

0.01

270.

373

0.05

260.

119

Wal

dp-

valu

e(J

oin

QE

)0.

0000

0.01

520.

2558

0.00

37

Note

s:T

his

tabl

ere

por

tspa

nelre

gres

sion

resu

lts

for

the

mod

elde

scri

bed

ineq

uati

on(1

)fo

rfu

ndch

arac

teri

stic

s.W

esu

btra

ctfr

omth

eva

riab

les

the

mea

nof

the

per

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


PSPP assets during our sample. Regression results pertaining tonon-PSPP-holding funds categorized as bond, mixed, and “other”are displayed in tables 8, 9, and 10, respectively. In this specifica-tion we exclude interactions between fund characteristics and QEdummies (again we focus on the regression explaining net purchasesrather than the value of holdings).

In the case of non-PSPP-holding bond funds, table 8, the firstresult that confirms a rebalancing channel is the significant coef-ficient on the QE second dummy in the case of Bondsgov . Thisshows that there were significantly higher net purchases of govern-ment bonds by non-PSPP-holding funds in the second phase of theEAPP (recall that these are funds that never held PSPP assetsand it therefore implies a rebalancing to government bonds thatwere not eligible for purchase under the purchase program—eithernon-EA government bonds or those with non-eligible maturities).The only other significant effect was the selling of bonds issued bydeposit-taking corporations in the QE anticipation period. This runscontrary to the presumed rebalancing channel.

Table 9 shows the analogous results for the mixed funds inthe non-PSPP sample. In this case we find evidence of rebalancingduring the anticipation period away from government-issued bonds(these could be bonds that are close substitutes for those anticipatedto be purchased under EAPP) towards bonds issued by the NFCs inboth the anticipation and first QE periods. The case of funds classi-fied as “others” is contained in table 10, and here we see that thereis a significant move away from equities and derivatives towards thenet purchases of bonds issued by NFCs in the second QE period andtowards bonds issued by unclassified issuers in both QE periods.

Once again, there is a possibility that investors are rebalanc-ing across funds of different types. To assess this, we again runregressions explaining the redemption and issuance behavior of thenon-PSPP-holding funds. Table 11 concerns redemption, issuance,leverage, and turnover for the non-PSPP bond funds. It might beexpected that fund selection by investors would be visible throughits effects on issuance and redemptions of units. In fact we see noevidence of increased redemptions. In support of movement into non-PSPP investment funds, we see strong evidence of increased issuancein the second QE period. Overall, this probably reflects an increasein investment in the non-PSPP fund sector specializing in bonds


le8.

Non

-PSP

PSam

ple

:B

ond

Fund/N

etP

urc

has

es

Var

iable

sEqu

ity

othe

rsD

eriv

ativ

esBon

dgo

vBon

ddt

cBon

dnfc

Bon

dot

hers

QE

antic

0.16

20.

0661

0.46

6−

0.36

1∗∗0.

139

−0.

103

(0.2

84)

(0.0

566)

(0.3

62)

(0.1

64)

(0.2

54)

(0.4

43)

QE

first

0.32

9−

0.07

260.

596

−0.

352

0.44

00.

148

(0.3

22)

(0.0

927)

(0.4

61)

(0.2

22)

(0.3

50)

(0.5

51)

QE

seco

nd0.

891

−0.

0864

1.63

5∗∗0.

0616

0.75

31.

246

(0.5

64)

(0.1

68)

(0.8

31)

(0.3

17)

(0.5

47)

(0.8

39)

Deb

tne

t0.

0001

010.

0003

51∗

0.00

200

0.00

118∗

0.00

0805

0.00

124

(0.0

0108

)(0

.000

206)

(0.0

0156

)(0

.000

606)

(0.0

0099

5)(0

.001

45)

US

long

yiel

d0.

912

−0.

115

1.90

2∗∗0.

284

0.73

51.

395∗

(0.5

74)

(0.1

75)

(0.8

33)

(0.3

18)

(0.5

23)

(0.7

99)

usco

rp-s

prea

d20.

244∗

−0.

0923

∗∗−

0.19

80.

0290

0.28

2∗0.

556∗∗

∗

(0.1

44)

(0.0

466)

(0.2

02)

(0.0

912)

(0.1

53)

(0.1

86)

usvi

xvo

l0.

0414

−0.

0160

−0.

0017

80.

0210

−0.

0025

90.

0791

∗

(0.0

355)

(0.0

100)

(0.0

487)

(0.0

186)

(0.0

338)

(0.0

414)

L.d

ep−

0.20

0∗∗∗

−0.

0828

−0.

0909

−0.

176∗∗

∗−

0.15

8∗∗∗

0.00

761

(0.0

653)

(0.0

845)

(0.0

570)

(0.0

607)

(0.0

495)

(0.0

557)

Obs

erva

tion

s2,

920

2,89

72,

814

2,85

02,

740

2,78

8N

umbe

rof

frfu

ndke

y37

737

837

637

737

637

4A

R1

p-va

lue

0.02

000.

0021

65.

68e-

061.

13e-

055.

21e-

081.

21e-

08A

R2

p-va

lue

0.11

20.

761

0.01

020.

232

0.00

525

0.01

24H

anse

n-J

p-va

lue

0.30

30.

174

0.01

470.

0165

0.00

0436

0.00

0323

Wal

dp-

valu

e(J

oin

QE

)0.

2699

0.72

190.

1406

0.01

320.

3801

0.03

22

Note

s:T

his

tabl

ere

por

tspa

nelr

egre

ssio

nre

sult

sfo

rth

em

odel

desc

ribed

ineq

uati

on(1

)fo

rne

tpu

rcha

ses.

We

repor

ton

lyth

ere

sult

sw

here

prop

orti

onal

hold

ings

are

onav

erag

eab

ove

4per

cent

inth

eP

SPP

sam

ple.

We

subt

ract

from

the

vari

able

sth

em

ean

ofth

eper

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.D

ebtne

tis

inbi

llion

sof

euro

s.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.

32 International Journal of Central Banking December 2019Tab

le9.

Non

-PSP

PSam

ple

:M

ixed

Fund/N

etP

urc

has

es

Var

iable

sEqu

ity

othe

rsD

eriv

ativ

esBon

dgo

vBon

ddt

cBon

dnfc

Bon

dot

hers

QE

antic

−1.

419∗∗

−0.

0545

−0.

280∗

−0.

0376

0.21

0∗∗∗

−0.

0129

(0.7

23)

(0.1

25)

(0.1

65)

(0.0

621)

(0.0

639)

(0.1

01)

QE

first

−0.

161

−0.

0329

−0.

332

0.03

570.

236∗∗

∗0.

0306

(1.2

01)

(0.1

41)

(0.2

93)

(0.0

821)

(0.0

756)

(0.1

36)

QE

seco

nd−

0.13

2−

0.07

00−

0.43

20.

0691

0.22

90.

0747

(1.8

74)

(0.2

40)

(0.5

37)

(0.1

08)

(0.1

42)

(0.1

44)

Deb

tne

t−

0.00

357

−5.

70e-

050.

0002

630.

0001

400.

0003

560.

0003

78(0

.003

17)

(0.0

0043

0)(0

.000

807)

(0.0

0024

4)(0

.000

249)

(0.0

0040

5)U

Slo

ngyi

eld

2.86

00.

0054

6−

0.30

00.

0791

0.17

30.

0717

(1.8

04)

(0.2

58)

(0.5

16)

(0.0

960)

(0.1

33)

(0.0

814)

usco

rpsp

read

21.

503∗∗

∗−

0.04

57−

0.14

00.

0266

0.01

910.

0520

(0.4

80)

(0.0

691)

(0.1

40)

(0.0

321)

(0.0

246)

(0.0

585)

usvi

xvo

l0.

250∗∗

∗−

0.01

580.

0026

6−

0.00

0124

−0.

0139

0.00

810

(0.0

889)

(0.0

153)

(0.0

237)

(0.0

0277

)(0

.009

41)

(0.0

118)

L.d

ep−

0.09

31∗

−0.

101∗

−0.

143∗

−0.

0668

−0.

185∗∗

−0.

236∗∗

∗

(0.0

524)

(0.0

586)

(0.0

781)

(0.1

13)

(0.0

749)

(0.0

780)

Obs

erva

tion

s3,

006

3,01

93,

022

3,02

13,

023

3,01

6N

umbe

rof

frfu

ndke

y41

441

541

641

441

441

6A

R1

p-va

lue

8.08

e-08

5.42

e-05

0.00

158

0.04

930.

0049

60.

0134

AR

2p-

valu

e0.

503

0.13

00.

0264

0.01

910.

148

0.12

3H

anse

n-J

p-va

lue

0.13

30.

112

0.03

440.

431

0.10

20.

204

Wal

dp-

valu

e(J

oin

QE

)0.

9879

0.95

820.

5131

0.75

110.

0053

0.77

31

Note

s:T

his

tabl

ere

por

tspa

nelr

egre

ssio

nre

sult

sfo

rth

em

odel

desc

ribed

ineq

uati

on(1

)fo

rne

tpu

rcha

ses.

We

repor

ton

lyth

ere

sult

sw

here

prop

orti

onal

hold

ings

are

onav

erag

eab

ove

4per

cent

inth

eP

SPP

sam

ple.

We

subt

ract

from

the

vari

able

sth

em

ean

ofth

eper

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.D

ebtne

tis

inbi

llion

sof

euro

s.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


le10

.N

on-P

SP

PSam

ple

:O

ther

Funds/

Net

Purc

has

es

Var

iable

sEqu

ity

othe

rsD

eriv

ativ

esBon

dgo

vBon

ddt

cBon

dnfc

Bon

dot

hers

QE

antic

−0.

553

0.03

930.

264

0.06

33−

0.01

22−

0.01

13(0

.940

)(0

.185

)(0

.367

)(0

.051

4)(0

.033

0)(0

.060

8)Q

Efir

st−

2.81

8∗∗−

0.37

1∗∗−

0.01

630.

0781

0.01

820.

224∗∗

(1.1

27)

(0.1

44)

(0.4

32)

(0.0

613)

(0.0

234)

(0.0

992)

QE

seco

nd−

3.75

5∗∗−

0.55

4∗∗−

0.33

70.

115

0.06

32∗

0.51

2∗∗∗

(1.6

37)

(0.2

29)

(0.6

67)

(0.1

05)

(0.0

347)

(0.1

95)

Deb

tne

t−

0.00

159

−0.

0006

960.

0004

470.

0001

650.

0001

080.

0003

95(0

.002

40)

(0.0

0046

2)(0

.001

24)

(0.0

0017

2)(7

.72e

-05)

(0.0

0037

2)U

Slo

ngyi

eld

−2.

061

−0.

356

−0.

721

0.07

230.

0735

0.51

9∗∗

(1.3

67)

(0.2

24)

(0.6

66)

(0.0

746)

(0.0

449)

(0.2

04)

usco

rpsp

read

2−

0.46

0−

0.11

3∗∗−

0.13

4−

0.00

400

0.00

332

0.04

87(0

.395

)(0

.056

0)(0

.183

)(0

.007

70)

(0.0

109)

(0.0

361)

usvi

xvo

l−

0.07

54−

0.02

29∗

−0.

0257

−0.

0029

4−

0.00

0889

0.01

34(0

.080

6)(0

.013

8)(0

.034

8)(0

.003

19)

(0.0

0292

)(0

.010

5)L.d

ep−

0.04

980.

0910

∗−

0.10

9∗0.

136

−0.

129

−0.

143∗∗

(0.0

333)

(0.0

489)

(0.0

564)

(0.1

08)

(0.0

940)

(0.0

670)

Obs

erva

tion

s2,

586

2,58

42,

509

2,56

12,

582

2,58

8N

umbe

rof

frfu

ndke

y31

931

931

831

631

831

9A

R1

p-va

lue

3.21

e-07

4.55

e-05

0.00

0182

0.09

130.

0442

0.02

19A

R2

p-va

lue

0.88

30.

0574

0.57

20.

353

0.62

60.

282

Han

sen-

Jp-

valu

e7.

40e-

050.

158

0.02

900

0.44

90.

389

Wal

dp-

valu

e(J

oin

QE

)0.

0400

0.03

590.

6919

0.40

500.

1893

0.03

09

Note

s:T

his

tabl

ere

por

tspa

nelr

egre

ssio

nre

sult

sfo

rth

em

odel

desc

ribed

ineq

uati

on(1

)fo

rne

tpu

rcha

ses.

We

repor

ton

lyth

ere

sult

sw

here

prop

orti

onal

hold

ings

are

onav

erag

eab

ove

4per

cent

inth

eP

SPP

sam

ple.

We

subt

ract

from

the

vari

able

sth

em

ean

ofth

eper

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.D

ebtne

tis

inbi

llion

sof

euro

s.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


le11

.N

on-P

SP

PSam

ple

:B

ond

Fund/F

und

Char

acte

rist

ics

Var

iable

sre

dem

nav

issu

ance

nav

leve

rage

nav

turn

over

nav

QE

antic

−0.

0006

75−

0.52

7−

1.14

4∗∗

0.02

73(0

.315

)(0

.419

)(0

.502

)(0

.111

)Q

Efir

st−

0.37

5−

0.25

5−

0.55

7−

0.11

2(0

.380

)(0

.449

)(0

.663

)(0

.128

)Q

Ese

cond

−0.

667

1.56

2∗∗

−0.

736

0.14

0(0

.671

)(0

.672

)(1

.070

)(0

.213

)D

ebtne

t−

0.00

0569

0.00

137

0.00

494∗

∗∗0.

0012

7∗∗∗

(0.0

0114

)(0

.001

27)

(0.0

0176

)(0

.000

325)

US

long

yiel

d−

0.91

62.

101∗

∗∗−

0.49

3−

0.01

11(0

.696

)(0

.664

)(1

.076

)(0

.211

)us

corp

spre

ad2

−0.

0337

1.09

1∗∗∗

−0.

0994

−0.

0471

(0.1

81)

(0.1

82)

(0.2

03)

(0.0

550)

usvi

xvo

l0.

111∗

∗0.

197∗

∗∗0.

0462

0.02

18∗

(0.0

454)

(0.0

525)

(0.0

767)

(0.0

127)

L.red

emna

v0.

0137

−0.

0259

0.24

8∗−

0.05

65(0

.042

1)(0

.029

7)(0

.143

)(0

.054

2)

Obs

erva

tion

s2,

932

2,91

42,

866

2,93

4N

umbe

rof

frfu

ndke

y37

937

737

837

9A

R1

p-va

lue

3.42

e-08

1.69

e-07

0.04

641.

71e-

06A

R2

p-va

lue

0.37

70.

371

0.49

70.

552

Han

sen-

Jp-

valu

e0.

0472

0.08

190.

0474

0.01

27W

ald

p-va

lue

(Joi

nQ

E)

0.58

250.

0000

0.69

920.

0547

Note

s:T

his

tabl

ere

por

tspa

nelre

gres

sion

resu

lts

for

the

mod

elde

scri

bed

ineq

uati

on(1

)fo

rfu

ndch

arac

teri

stic

s.W

esu

btra

ctfr

omth

eva

riab

les

the

mea

nof

the

per

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


that can give investors similar exposure to what they would usuallyobtain from investment in units of PSPP-holding funds.

There is also a significant increase in issuance in both the firstand second QE periods for mixed funds (table 12), combined witha distinct absence of any evidence for increased redemptions. Sothis category of non-PSPP holders also experienced net new invest-ment. For “other” funds we find no evidence of redemption asso-ciated with QE but also no net issuance (table 13). Overall, thereis substantial evidence that investors have moved their investmentsaway from PSPP-holding funds towards non-PSPP-holding funds,and this confirms an investor-mediated rebalancing channel. This isalso subject to a caveat, to which we now return, that such fundscould be redirecting the new investment to non-EA assets.

5.2.1 Rebalancing: Euro Area and the Rest of the World(RoW)

The results so far concern the rebalancing of portfolios by asset type.Figure 7, panels A–D, shows four different views of net purchasesaccording to the region of issuance of the assets concerned. Figure7A shows, for the case of PSPP-holding funds, the EA/RoW break-down of the two asset categories where significant parameters on netpurchases were found (see table 5, specifically the columns relatingto government bonds and bonds issued by deposit-taking corpora-tions). In all eleven quarters shown there is strong evidence of netselling of EA government bonds. However, in nine out of eleven ofthe quarters we also observe net purchases of government bonds thatare issued outside the EA. The situation concerning net purchasesof bonds issued by deposit-taking corporations suggests that thisactivity was mostly in favor of those issued outside the EA.

Figure 7B presents the EA/RoW breakdown of the net purchasesof government bonds by non-PSPP holders (these were also found tobe statistically significant before being categorized as EA or RoW).The graphical breakdown clearly indicates that the majority of therebalancing by non-PSPP holders was towards government bondsissued outside the EA. Likewise, figures 7C and 7D show that a largeproportion of the “mixed” and “other” non-PSPP funds also exhibitrebalancing towards RoW assets. The associated regressions (notshown) support this finding once fund characteristics are interacted


le12

.N

on-P

SP

PSam

ple

:M

ixed

Fund/F

und

Char

acte

rist

ics

Var

iable

sre

dem

nav

issu

ance

nav

leve

rage

nav

turn

over

nav

QE

antic

0.34

60.

408

−0.

278

0.05

94(0

.265

)(0

.270

)(0

.425

)(0

.094

9)Q

Efir

st0.

193

0.71

5∗∗

−0.

0826

0.04

67(0

.296

)(0

.327

)(0

.548

)(0

.119

)Q

Ese

cond

0.66

80.

988∗

0.97

40.

0970

(0.4

32)

(0.5

25)

(1.0

33)

(0.1

53)

Deb

tne

t0.

0020

0∗∗

0.00

215∗

∗0.

0015

20.

0007

16∗∗

∗

(0.0

0086

3)(0

.000

920)

(0.0

0209

)(0

.000

278)

US

long

yiel

d0.

0254

1.66

2∗∗∗

−0.

272

−0.

182

(0.4

43)

(0.5

85)

(1.1

28)

(0.1

73)

usco

rpsp

read

2−

0.00

190

0.43

5∗∗∗

0.08

410.

0902

∗

(0.1

28)

(0.1

28)

(0.2

06)

(0.0

469)

usvi

xvo

l0.

0104

0.01

540.

0358

0.00

530

(0.0

256)

(0.0

305)

(0.0

426)

(0.0

0816

)L.d

epna

v−

0.11

5∗∗

−0.

0389

∗0.

346∗

∗−

0.17

0∗∗∗

(0.0

478)

(0.0

203)

(0.1

77)

(0.0

566)

Obs

erva

tion

s3,

051

3,04

73,

030

3,05

9N

umbe

rof

frfu

ndke

y41

541

541

441

6A

R1

p-va

lue

2.73

e-05

6.43

e-09

0.07

240.

0038

7A

R2

p-va

lue

0.00

672

0.35

60.

595

0.10

6H

anse

n-J

p-va

lue

0.03

970.

456

0.15

90.

0143

Wal

dp-

valu

e(J

oin

QE

)0.

1908

0.09

060.

2102

0.79

90

Note

s:T

his

tabl

ere

por

tspa

nelre

gres

sion

resu

lts

for

the

mod

elde

scri

bed

ineq

uati

on(1

)fo

rfu

ndch

arac

teri

stic

s.W

esu

btra

ctfr

omth

eva

riab

les

the

mea

nof

the

per

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


le13

.N

on-P

SP

PSam

ple

:O

ther

Funds/

Fund

Char

acte

rist

ics

Var

iable

sre

dem

nav

issu

ance

nav

leve

rage

nav

turn

over

nav

QE

antic

0.55

3∗∗

0.15

6−

8.09

5∗∗∗

0.18

2∗

(0.2

78)

(0.4

00)

(2.4

63)

(0.0

932)

QE

first

0.05

240.

0546

−8.

426∗

∗∗0.

281∗

∗

(0.4

26)

(0.4

03)

(3.1

93)

(0.1

22)

QE

seco

nd−

0.34

60.

499

−25

.13∗

∗∗0.

259

(0.6

65)

(0.6

28)

(6.6

27)

(0.1

61)

Deb

tne

t0.

0009

150.

0023

7∗−

0.02

08∗∗

0.00

0202

(0.0

0083

6)(0

.001

36)

(0.0

0968

)(0

.000

288)

US

long

yiel

d−

1.65

4∗∗∗

0.61

1−

23.4

7∗∗∗

−0.

118

(0.6

29)

(0.7

54)

(5.1

76)

(0.1

70)

usco

rpsp

read

20.

0426

0.52

1∗∗∗

−3.

492∗

∗0.

0373

(0.1

89)

(0.1

53)

(1.4

21)

(0.0

491)

usvi

xvo

l−

0.01

040.

0426

0.70

7∗∗

0.02

09∗

(0.0

340)

(0.0

324)

(0.3

19)

(0.0

109)

L.d

ep−

0.03

85−

0.05

390.

568∗

∗∗−

0.17

7∗∗∗

(0.0

459)

(0.0

392)

(0.0

452)

(0.0

572)

Obs

erva

tion

s2,

617

2,61

12,

474

2,61

9N

umbe

rof

frfu

ndke

y31

931

831

331

9A

R1

p-va

lue

2.37

e-09

4.12

e-07

3.74

e-07

0.00

0192

AR

2p-

valu

e0.

706

0.68

50.

0104

0.80

2H

anse

n-J

p-va

lue

0.00

254

0.00

0329

2.54

e-05

0.02

28W

ald

p-va

lue

(Joi

nQ

E)

0.51

150.

4594

0.00

010.

0693

Note

s:T

his

tabl

ere

por

tspa

nelre

gres

sion

resu

lts

for

the

mod

elde

scri

bed

ineq

uati

on(1

)fo

rfu

ndch

arac

teri

stic

s.W

esu

btra

ctfr

omth

eva

riab

les

the

mea

nof

the

per

iod

2014

:Q1–

2014

:Q3.

Coe

ffici

ent

esti

mat

esar

esc

aled

by10

0.St

anda

rder

rors

are

inpa

rent

hese

s.**

*,**

,an

d*

deno

tep

<0.

01,p

<0.

05,an

dp

<0.

1,re

spec

tive

ly.


Figure 7. Asset Type and Area of Issuance

A. PSPP Holders B. Non-PSPP Holders: Bond Funds

C. Non-PSPP Holders: Mixed Funds D. Non-PSPP Holders: Other Funds

with QE dummies. In particular, funds with relatively large issuanceof new units significantly rebalance their portfolios away from EA-issued assets. Overall, there appears to be substantial evidence forrebalancing that may trigger an exchange rate adjustment channel(i.e., depreciation of the euro and increased imported inflation).

5.2.2 Rebalancing by Maturity

Table 14 shows panel regression results pertaining to the rebalancingacross assets with different maturities for the case of PSPP-holdinginvestment funds. This again shows a significantly negative parame-ter on the second QE dummy variable for the case of maturities cov-ered by the program (maturity greater than two years). We see thatthere is statistically significant evidence of a move towards bondswith maturity less than two years in the first QE period.

In the case of non-PSPP-holding funds, we found evidence ofrebalancing effects as follows. For the bond funds we found a con-sistently significant increase in holdings of government bonds (these


Table 14. PSPP Sample: Net Purchases/Maturity

Variables less1year less2year more2year

QE antic −1.316∗∗ 0.259∗ −0.626(0.619) (0.133) (1.020)

QE first 0.0804 0.479∗∗ −1.051(0.947) (0.232) (1.410)

QE second −0.840 0.369 −3.577∗

(1.679) (0.268) (1.991)Debt net −0.00140 −0.000278 −0.00944∗∗

(0.00250) (0.000477) (0.00372)US long yield −0.944 0.466∗ −0.572

(1.503) (0.252) (2.009)us corp spread2 0.559∗∗ 0.237∗∗∗ 0.112

(0.243) (0.0726) (0.657)us vix vol 0.0847 −0.0164 −0.0492

(0.0718) (0.0167) (0.133)L.dep −0.295∗∗∗ −0.247∗∗∗ −0.0873

(0.0652) (0.0430) (0.0800)

Observations 1,338 1,338 1,343Number of frfundkey 174 174 174AR1 p-value 0.00131 0.00161 0.000172AR2 p-value 0.0155 0.0891 0.0866Hansen-J p-value 0.177 0.0339 0.281Wald p-value 1.24e-07 4.99e-10 0.000359

Notes: This table reports panel regression results for the model described in equa-tion (2) for net purchases. We subtract from the variables the mean of the period2014:Q1–2014:Q3. Coefficient estimates are scaled by 100. Standard errors are inparentheses. ***, **, and * denote p < 0.01, p < 0.05, and p < 0.1, respectively.

are non-PSPP bonds by definition) with maturity greater than twoyears. For funds classed as “other” we also find a positive and signif-icant increase in holdings of assets with maturity greater than twoyears. These results are supportive of QE effects working throughinvestor preferences for assets with greater average duration (andsome additional yield). This should contribute to a flattening of theterm structure and an improvement in long-term funding costs forboth the public and private sectors. Hedge, mixed, and real estatefunds never provide significant results.


6. Robustness Analysis

A number of robustness checks were carried out. Firstly, we exam-ined a number of alterations in the lag structure of the Arellano-Bover GMM specification. We found that results were stable for theuse of lags as instruments for the case of (i) lags 2 and 3, (ii) lags2, 3, and 4, or (iii) lags 2 to 5. In many cases there is actuallyquite weak evidence for the need for a lagged dependent variable inthe regression. This suggests that a static panel could be used asan alternative to the GMM approach. We found that (for the basicmodel without the interaction with fund characteristics) many ofthe core results discussed above survive when we move to a staticfixed-effects panel regression method. Specifically, the evidence of amove by PSPP holders from PSPP assets to those issued by deposit-taking corporates remains significant. We had some concern that ourPSPP sample of funds could contain many funds that are strictlymandated to only invest in PSPP assets. However, even when wedropped funds with consistently over 90 percent of their portfolio inEA-issued government bonds, the results remained broadly similar(i.e., coefficients changed slightly but signs and significance remainedthe same).

As regards the sample of funds that had been dropped from themain analysis due to extremely high daily turnover, we carried outall of the same regressions as discussed above on this sample toensure that we were not losing important information. We foundsome similar results for such funds but, more often than not, theresults lacked consistent statistical significance. This reflects the factthat the sample of funds in this category is quite small, is diversein behaviors, and the number of parameters being estimated is rel-atively large (especially in the specifications that includes interac-tion terms). There is however evidence of a statistically significantincrease in investment in Bonds other in the standard regressionwithout interaction terms for the PSPP-holding high-turnover fundsin anticipation of QE and, even more so, for the first and second QEphases. Together with the negative sign (albeit statistically insignif-icant) on all the QE dummies for Bonds gov , this can be regardedas valid evidence for a rebalancing from PSPP assets to non-PSPPassets. PSPP-holding high-turnover funds also tend to show a movetowards holding short-term securities (this is into securities outside


the purchasable basket). The results for the non-PSPP high-turnoverfunds do not provide significant evidence of rebalancing, but in thiscase there is some evidence of increased redemptions and decreasedissuance of fund units across the QE periods.

7. Conclusion

We have examined the portfolio rebalancing behavior of a large sam-ple of investment funds before and during the Eurosystem’s PublicSector Purchase Programme (PSPP). We find evidence that fundsmost exposed to the purchase program rebalanced their portfoliostowards bonds issued by deposit-taking corporations, into assetswith maturities outside the maturities eligible for purchase underthe program, and away from euro-denominated assets. For subsetsof funds focused on non-PSPP eligible assets, we found evidence ofmoves into longer-term securities and away from euro-denominatedassets.

Via an examination of redemptions and issues by funds of theirown shares (units), we found that investors rebalanced away fromPSPP-holding funds towards those focused on holding other types ofassets. This probably reflects an increase in investment in the non-PSPP fund sector, those specializing in bonds that can give investorssimilar exposure to what they would usually obtain from investmentin units of PSPP-holding funds. Investor rebalancing behavior inter-acts significantly with portfolio rebalancing behavior of the fundsthemselves.

Overall, our analysis provides support for rebalancing effects,but it seems that these are mostly towards assets issued outside theeuro area. In economic terms (based on the size of net purchases), theeffects do not appear large. Given that we only find significant effectsafter the pace of purchases was increased, the choice of program scalemay be crucial. There is no evidence of rebalancing towards equi-ties, which supports the views of Ball et al. (2017) that extendingthe purchase program in this direction may be beneficial. At thetime of the announcement of the purchase program, equities rosesignificantly, but it has been shown that this was due mostly to aperceived lower discount rate over an extended future horizon. Theassociated drop in the cost of capital should have produced issuanceof new equity. But we see no sign of this in fund portfolios. This


sector, during the sample, may have been holding off until issuanceconditions improved. The same could be said of banks, since wealso find little evidence of rebalancing towards EA bonds or equitiesissued by the deposit-taking corporations.

As with equity valuation, it might be expected that our findingsof continued portfolio rebalancing towards non-EA assets may havehelped to generate euro weakness. We know however, from otherresearch, that the bulk of euro weakening over this entire period canbe linked to the initial announcement of the program itself. Contin-ued flow towards non-EA assets during the sample period, despiteeuro weakness, suggests a reluctance on the part of investors (bothfunds and fund investors) to bet on growth prospects given perceivedrisks.

Our sample period ends in the third quarter of 2016. Recentdevelopments, visible from sector-based asset holdings, seem to indi-cate some strengthening of effects from the asset purchase program.These include (i) an improvement in net external assets, (ii) a rebal-ancing towards European equities, and (iii) a significant increase ininvestors’ preferences for debt securities issued by euro-area banks.Whether these developments are attributable particularly to assetpurchases is open to question, and this therefore remains an impor-tant avenue for a possible future extension of the current analysis.

References

Albertazzi, U., B. Becker, and M. Boucinha. 2016. “Portfolio Rebal-ancing and the Transmission of Large-Scale Asset Programs:Evidence from the Euro Area.” Working Paper (November).

Altavilla, C., M. Boucinha, and J.-L. Peydro. 2017. “Monetary Pol-icy and Bank Profitability in a Low Interest Rate Environment.”ECB Working Paper No. 2105 (October).

Altavilla, C., F. Canova, and M. Ciccarelli. 2016. “Mending theBroken Link: Heterogeneous Bank Lending and Monetary PolicyPass-through.” ECB Working Paper No. 1978 (November).

Altavilla, C., G. Carboni, and R. Motto. 2015. “Asset Purchase Pro-grammes and Financial Markets: Lessons from the Euro Area.”ECB Working Paper No. 1864.


Altavilla, C., M. Pagano, and S. Simonelli. 2017. “Bank Exposuresand Sovereign Stress Transmission.” Review of Finance 21 (6):2103–39.

Andrade, P., J. Breckenfelder, F. De Fiore, P. Karadi, and O. Tris-tani. 2016. “The ECB’s Asset Purchase Programme: An EarlyAssessment.” ECB Working Paper No. 1956 (September).

Andrade, P., C. Cahn, H. Fraisse, and J.-S. Mesonnier. 2017.“Can the Provision of Long-Term Liquidity Help to Avoid aCredit Crunch? Evidence from the Eurosystem’s LTRO.” Work-ing Paper.

Arellano, M., and S. Bond. 1991. “Some Tests of Specificationfor Panel Data: Monte Carlo Evidence and an Application toEmployment Equations.” Review of Economic Studies 58 (2):277–97.

Arellano, M., and O. Bover. 1995. “Another Look at the Instrumen-tal Variable Estimation of Error-Components Models.” Journalof Econometrics 68 (1): 29–51.

Bai, J., and S. Ng. 2010. “Instrumental Variables Estimation ina Data Rich Environment.” Econometric Theory 26 (6): 1577–1606.

Ball, L., J. Gagnon, P. Honohan, and S. Krogstrup. 2017. “WhatElse Can Central Banks Do?” CEPR & ICMB Geneva Reportson the World Economy No. 18.

Baum, C., M. E. Schaffer, and S. Stillman. 2003. “Instrumental Vari-ables and GMM: Estimation and Testing.” Stata Journal 3 (1):1–31.

Baumeister, C., and L. Benati. 2013. “Unconventional MonetaryPolicy and the Great Recession: Estimating the MacroeconomicEffects of a Spread Compression at the Zero Lower Bound.”International Journal of Central Banking 9 (2): 165–211.

Benford, J., S. Berry, K. Nikolov, M. Robson, and C. Young. 2009.“Quantitative Easing.” Quarterly Bulletin (Bank of England) Q2(June 12): 90–100.

Bergant, K. 2017. “Quantitative Easing and Portfolio Rebalancing:Micro Evidence from Irish Resident Banks.” Economic LetterNo. 7, Central Bank of Ireland.

Blundell, R., and S. Bond. 1998. “Initial Conditions and MomentRestrictions in Dynamic Panel Data Models.” Journal of Econo-metrics 87 (1): 115–43.


Bollerslev, T., G. Tauchen, and H. Zhou. 2009. “Expected StockReturns and Variance Risk Premia.” Review of Financial Studies22 (11): 4463–92.

Brunnermeier, M. K., and Y. Sannikov. 2014. “A MacroeconomicModel with a Financial Sector.” American Economic Review 104(2): 379–421.

Bubeck, J., M. M. Habib, and S. Manganelli. 2017. “The Port-folio of Euro Area Fund Investors and ECB Monetary PolicyAnnouncements.” ECB Working Paper No. 2116 (December).

Carpenter, S., S. Demiralp, J. Ihrig, and E. Klee. 2013. “Analyz-ing Federal Reserve Asset Purchases: From Whom Does the FedBuy?” Finance and Economics Discussion Series No. 2013-32,Board of Governors of the Federal Reserve System.

———. 2015. “Analyzing Federal Reserve Asset Purchases: FromWhom Does the Fed Buy?” Journal of Banking and Finance 52(March): 230–44.

Carpinelli, L., and M. Crosignani. 2017. “The Effect of Central BankLiquidity Injections on Bank Credit Supply.” Finance and Eco-nomics Discussion Series No. 2017-038, Board of Governors ofthe Federal Reserve System.

Christensen, J., and G. Rudebusch. 2012. “The Response of InterestRates to US and UK Quantitative Easing.” Economic Journal122 (564): F385–F414.

Coimbra, N., and H. Rey. 2016. “Financial Cycles with Heteroge-neous Intermediaries.” Working Paper, London Business School.

D’Amico, S., W. English, D. Lopez-Salido, and E. Nelson. 2012.“The Federal Reserve’s Large-Scale Asset Purchase Programs:Rationale and Effects.” Economic Journal 122 (564): F415–F446.

D’Amico, S., and T. King. 2013. “Flow and Stock Effects of Large-Scale Treasury Purchases: Evidence on the Importance of LocalSupply.” Journal of Financial Economics 108 (2): 425–48.

Demertzis, M., and G. B. Wolff. 2016. “The Effectiveness of theEuropean Central Bank’s Asset Purchase Programme.” BruegelPolicy Contribution.

DeSantis, R. A., and F. Holm-Hadulla. 2017. “Flow Effects of Cen-tral Bank Asset Purchases on Euro Area Sovereign Bond Yields:Evidence from a Natural Experiment.” ECB Working Paper No.2052.


Deutsche Bundesbank. 2017. “The Eurosystem’s Bond Purchasesand the Exchange Rate of the Euro.” Monthly Report 69 (1):13–39.

Eggertsson, G. B., and M. Woodford. 2003. “The Zero Bound onInterest Rates and Optimal Monetary Policy.” Brookings Paperson Economic Activity 34 (1): 139–211.

Engen, E. M., T. Laubach, and D. Reifschneider. 2015. “The Macro-economic Effects of the Federal Reserve’s Unconventional Mon-etary Policies.” Finance and Economics Discussion Series No.2015-005, Board of Governors of the Federal Reserve System.

Eser, F., and B. Schwaab. 2016. “Evaluating the Impact of Uncon-ventional Monetary Policy Measures: Empirical Evidence fromthe ECB’s Securities Markets Programme.” Journal of FinancialEconomics 119 (1): 147–67.

European Central Bank. 2013. “Box 3: Early Repayment of FundsRaised through Three-year Longer-term Refinancing Operations:Economic Rationale and Impact on the Money Market.” Eco-nomic Bulletin (February): 30–31.

———. 2015. “Box 2: Recent Equity Price Developments in the EuroArea and the United States.” Economic Bulletin 4 (January):34–38.

Fratzscher, M., M. Lo Duca, and R. Straub. 2016. “ECB Uncon-ventional Monetary Policy: Market Impact and InternationalSpillovers.” IMF Economic Review 64 (1): 36–74.

Gagnon, J., M. Raskin, J. Remache, and B. Sack. 2010. “Large-scaleAsset Purchases by the Federal Reserve: Did They Work?” StaffReport No. 411, Federal Reserve Bank of New York.

———. 2011. “The Financial Market Effects of the Federal Reserve’sLarge-scale Asset Purchases.” International Journal of CentralBanking 7 (1): 3–43.

Glick, R., and S. Leduc. 2013. “The Effects of Unconventional andConventional US Monetary Policy on the Dollar.” Working PaperNo. 2013-11, Federal Reserve Bank of San Francisco.

Haldane, A., M. Roberts-Sklar, T. Wieladek, and C. Young. 2016.“QE: The Story So Far.” CEPR Discussion Paper No. 11691.

Hau, H., and S. Lai. 2016. “Asset Allocation and Monetary Policy:Evidence from the Eurozone.” Journal of Financial Economics120 (2): 309–29.


Hogen, Y., and M. Saito. 2014. “Portfolio Rebalancing Following theBank of Japan’s Government Bond Purchases: Empirical Analy-sis Using Data on Bank Loans and Investment Flows.” ResearchPaper No. 14-06-19 (June), Bank of Japan.

Holtz-Eakin, D., W. Newey, and H. Rosen. 1988. “Estimating VectorAutoregressions with Panel Data.” Econometrica 56 (6): 1371–95.

Joyce, M., A. Lasaosa, I. Stevens, and M. Tonks. 2011. “The Finan-cial Market Impact of Quantitative Easing in the United King-dom.” International Journal of Central Banking 7 (3): 113–61.

Joyce, M., Z. Liu, and I. Tonks. 2015. “Institutional Investor Port-folio Allocation, Quantitative Easing and the Global FinancialCrisis.” Working Paper No. 510 (September), Bank of England.

Joyce, M., and M. Tonks. 2012. “QE and the Gilt Market: A Disag-gregated Analysis.” Economic Journal 122 (564): F348–F384.

Koijen, R. S. J., F. Koulischer, B. Nguyen, and M. Yogo. 2017.“Euro-area Quantitative Easing and Portfolio Rebalancing.”American Economic Review 107 (5): 621–27.

Krishnamurthy, A., and A. Vissing-Jorgensen. 2011. “The Effectsof Quantitative Easing on Interest Rates: Channels and Implica-tions for Policy.” Brookings Papers on Economic Activity 42 (2,Fall): 215–87.

Stein, J. C. 2014. “Incorporating Financial Stability Considerationsinto a Monetary Policy Framework.” Remarks at the Interna-tional Research Forum on Monetary Policy, Washington, DC,March 21.

Valiante, D. 2015. “The ‘Visible Hand’ of the ECB’s QuantitativeEasing.” CEPS Working Paper No. 407 (May).

Vayanos, D., and J.-L. Vila. 2009. “A Preferred-habitat Model ofthe Term Structure of Interest Rates.” Discussion Paper No.641, Financial Markets Group, London School of Economics andPolitical Science.

Weale, M., and T. Wieladek. 2014. “What Are the MacroeconomicEffects of Asset Purchases?” Discussion Paper No. 42, Bank ofEngland.

Windmeijer, F. 2005. “A Finite Sample Correction for the Vari-ance of Linear Efficient Two-step GMM Estimators.” Journal ofEconometrics 126 (1): 25–51.

Woodford, M. 2012. “Inflation Targeting and Financial Stability.”NBER Working Paper No. 17967.

Documents

The Portfolio Rebalancing Effects of the ECB's Asset ... · active rebalancing behavior as distinct from passive rebalancing as described in Bubeck, Habib, and Manganelli (2017) (this