Global Impact and Challenges of Unconventional Monetary Policies - Background Paper
Author: International Monetary Fund
Publication Date:
09 Mar 2013
eISBN:
9781498341349
Language:
English
Keywords:
PP; monetary policy; emerging market; exchange rate; banking sector; central bank; output gap; market capitalization; global bond; short-term debt
Download PDF (1.9 MB)

This paper provides case studies of 13 of the largest non-UMP countries. The case studies begin with an overview of recent macro-economic developments as well as capital flow patterns during the crisis up to the first U.S. tapering announcement in May 2013. Country experiences with capital inflows are judged along five dimensions: (i) the size of capital inflows, (ii) policies used to manage inflows, (iii) external stability, measured by exchange rate overvaluation and current account deficits relative to fundamentals,2 (iv) asset price and credit market reactions, and (v) financial sector stability. Case studies mostly draw on published IMF Staff Reports for each country, as well as the 2013 Pilot External Stability Report (IMF 2013d).

Abstract

This paper provides case studies of 13 of the largest non-UMP countries. The case studies begin with an overview of recent macro-economic developments as well as capital flow patterns during the crisis up to the first U.S. tapering announcement in May 2013. Country experiences with capital inflows are judged along five dimensions: (i) the size of capital inflows, (ii) policies used to manage inflows, (iii) external stability, measured by exchange rate overvaluation and current account deficits relative to fundamentals,2 (iv) asset price and credit market reactions, and (v) financial sector stability. Case studies mostly draw on published IMF Staff Reports for each country, as well as the 2013 Pilot External Stability Report (IMF 2013d).

Non-UMP Country Case Studies1

1. This section provides case studies of 13 of the largest non-UMP countries. The case studies begin with an overview of recent macro-economic developments as well as capital flow patterns during the crisis up to the first U.S. tapering announcement in May 2013. Country experiences with capital inflows are judged along five dimensions: (i) the size of capital inflows, (ii) policies used to manage inflows, (iii) external stability, measured by exchange rate overvaluation and current account deficits relative to fundamentals,2 (iv) asset price and credit market reactions, and (v) financial sector stability. Case studies mostly draw on published IMF Staff Reports for each country, as well as the 2013 Pilot External Stability Report (IMF 2013d).

2. The prospects for capital outflows draw on assessments of countries’ exposure and resilience. This is as explained in detail in the main paper (Box 7). Exposure measures the likelihood of market volatility and capital outflows following tapering in advanced economies (AEs) (in practice, in the United States (U.S.)). Resilience measures the ability of countries to withstand the pressures from potential market volatility and capital outflows. The exposure and resilience of countries is judged using the indicators described and explained in Table 1 and shown in Appendix Table 1.

Table 1.

Measures of Exposure and Resilience

article image
article image
Appendix Table 1.

Selected Indicators

(In millions of SDRs; as of September 3, 2010)

article image

Based on regression of non-UMP country 10-year bond yields on U.S. 10-year bond yields. The regression was run in changes over two-day intervals immediately following U.S. FOMC announcements between January 1, 2003 and May 20, 2013. Bolded numbers indicate statistically significant coefficients.

Cumulative change in EM bond yields using 2-day windows on 05/22/2013 and 06/19/2013.

Z-scores of average flows (equity and bond) in the month after tapering announcement on 5/22/2013 relative to flows since 2009. A z-score represents the deviation from the long-term average expressed in the number of standard deviations. Source: EPFR data and staff calculation.

Lowest between Moody’s and S&P ratings for local currency debt as of mid-August 2013 and expressed in S&P ratings. Source: Bloomberg.

Based on 2012 data. Source: World Bank WDI.

Market size of domestic investors that could provide funding in case of a sudden stop. Data as of end-December 2011 for South Africa; end-Jan 2012 for Brazil; end-March 2012 for Indonesia, Korea, Poland; end-April 2012 for Turkey; end-May 2012 for Mexico; and end-2009 for China and Thailand. Source: GFSR April 2012.

Foreign ownership of domestic portfolio equities as a percent of GDP. Based on 2012 data, except for Indonesia, South Africa and Thailand, which use data from 2011. Sources: WEO July 2013 and IFS.

Total Gross External Debt in 2012 Source: WEO July 2013

The higher the indicated primary balance adjustment, the greater the degree of fiscal tightening needed to reduce the debt-to-GDP ratio to 60 percent for AEs and 40 percent for EMs in 2030, and thus the less available fiscal space. Source: Fiscal Monitor, April 2013, Statistical Table 13.

This measure is calculated as the differences between WEO projected CPI inflation in 2013 and inflation target (upper bound is used if central bank targets at a range). For South Africa, core inflation target is used. Source: WEO July 2013.

Source: WEO July 2013

Differences between REER and those consistent with medium-term fundamentals and desirable policies. Assessments prepared in May 2013. Source: 2013 External Stability Report, IMF 2013d.

Based on 2012 data. Source: WEO July 2013.

Number of months of imports covered by FX reserves. Based on 2012 data. Source: Staff estimates and IFS.

Based on 2012 data from the Financial Soundness Indicators (FSI), except for Thailand, which is based on 2012 data from the Bank of Thailand.

3. Some caveats should be raised. First, the list of exposure and resilience indicators is by nature incomplete—it could be expanded—and different indicators are more or less appropriate for different countries. Second, economic developments are bound to change quickly and new evidence on countries’ exposure and resilience will continue to emerge. The assessment provided in this Background Paper only takes into account data through mid-August 2013 when possible. Thus, any conclusions are preliminary.

4. This said, non-UMP countries differ considerably in their measured exposure and resilience. The more developed non-UMP economies (Australia—higher resilience, as well as Canada and Korea—lower exposure) as well as other Emerging Market Economies (EMEs) with higher resilience and/or lower exposure are expected to fare relatively well following a U.S. exit. Other countries appear more vulnerable (due to both higher exposure and lower resilience), and some are borderline cases between these two extremes.

5. Evidence from the second and third indicators of “exposure” is illustrated below. It is noteworthy that countries showing the largest response to U.S. Federal Reserve announcements of bond purchases did not necessarily show the largest response to tapering announcements. This may be due to the changing nature of investor positioning and market liquidity.

Figure 1.
Figure 1.

Total Cumulated Changes in 10-year Bond Yields Following U.S. LSAP Announcements and Tapering Announcements

Citation: Policy Papers 2013, 013; 10.5089/9781498341349.007.A001

Source: Fund staff estimates.Note: Changes in yields are computed in the day following each announcement, then cumulated. Tapering announcements occurred on May 22 and June 19, 2013.
Figure 2.
Figure 2.

Measure of Capital Outflows Following U.S. Tapering Announcements

Citation: Policy Papers 2013, 013; 10.5089/9781498341349.007.A001

Sources: EPFR Global and Fund staff estimates.Note: Z-scores represent the difference of average capital flows (equity and bond) in the month after first U.S. tapering announcement (May 22, 2013) relative to flows since 2009, normalized by the standard deviation of flows.

Australia

article image

Brazil

article image

Canada

article image

China

article image

India

article image

Indonesia

article image

Korea

article image

Mexico

article image

Poland

article image

Russia

article image

South Africa

article image

Thailand

article image

Turkey

article image

Capital Flows and Financial Vulnerability in a Model of Macrofinancial Risk and Macroeconomic Stress4

6. The impact of capital flows and their reversals on the financial sector and the real economy can be nonlinear. Depending on the extent of the risks in the balance sheets of lenders and borrowers, a large-enough adverse shock can force the economy out of a corridor of stabilit y (where standard adjustment mechanisms make for a smooth and predictable return to the normal) into a region where global nonlinearities and asymmetries rapidly take over.

7. Standard macroeconomic models, by construction, are not capable of dealing with macrofinancial stress scenarios. These models typically overlook two sources of instability. First, they ignore the existence of endogenous aggregate (non-diversifiable) risks on the balance sheets of financial institutions and borrowers. Second, they are routinely solved by local approximation methods (such as linearization or higher-order approximation), and hence cannot provide any insights into global nonlinearities arising from such balance-sheet risks. Arguably, such models can only explain macroeconomic behavior within the corridors of stability; and attempts to extrapolate local dynamics to the regions of large distress inevitably result in an overly benign picture.

8. IMF staff has therefore developed a new type of model allowing for global nonlinear feedback in scenarios of large distress. The novelty lies in integrating several macrofinancial amplification mechanisms within a broader macroeconomic framework. The model endogenizes and interconnects the notions of aggregate credit risk, loan portfolio value of banks, bank capital buffers, and relative costs of internal and external equity flows. The output of the model is primarily illustrative; patterns are more telling than the exact numerical values of series. The model first and foremost provides a coherent analysis and understanding of the mechanisms at the heart of tail risk events and large distress episodes, and can help to guide the design of robust policies.

A. Brief Description of the Model

9. The model combines relatively standard macroeconomic assumptions based on optimizing behavior with two concepts from the finance and banking literature. The first is the existence of aggregate, non-diversifiable credit risk on the loan books of banks, with the risk dynamics derived endogenously from macroeconomic developments. The second is the optimal choice of bank balance sheets and capital buffers under uncertainty, constrained by the risk-bearing capacity of bank capital.

10. The real side of the model represents a small open economy in standard fashion, with production goods for local consumption and exports. The model can be parameterized to represent a variety of different types of open economies. The model distinguishes between the short-run and long-run elasticity of substitution in expenditure switching, and allows for different degrees of import substitution, as well as weights on permanent and current income in determining aggregate demand.

11. The financial sector consists of banks. Banks extend non-traded bank loans, and create matching liabilities in the form of bank deposits. They are also required to hold capital. The model distinguishes between saving and financing. Bank deposits, created at the moment of extending a new loan (as in the real world), are used to finance consumption, investment, and purchases of imports, or are accepted to finance nonresidents’ purchases of local exports.

12. Individual bank loans and overall loan portfolios are both risky. Debtors can default on their loans depending on the evolution of their income and wealth. The risks of individual loans are correlated: loans share a common, systemic component, determined by aggregate macroeconomic developments, ex-ante unpredictable. As a result, banks are able to diversify some risk (by extending loans to a large number of borrowers) but not all risks. Specifically, bank loan portfolios remain exposed to aggregate risk.

13. Lending is constrained by regulatory capital buffers. Given the amount of capital, banks optimize the size of their balance sheets by expanding or reducing their loan portfolios (and the amount of the matching bank deposits). Because of the existence of non-diversifiable risk, banks choose to hold capital in excess of regulatory requirements to minimize the cost of possible capital shortfalls in the future. The buffers vary endogenously over time in a pro-cyclical fashion, as they depend on the riskiness of the loans which in turn depend on the income and wealth of borrowers.

14. The banking sector adds a critical feedback mechanism to the model. Unforeseen adverse shocks cause a rise in impaired loans, and subsequent write-downs. Thinner capital buffers induce banks to liquidate some of their assets and increase the price of bank lending, triggering a vicious circle of fire sales and credit crunches.

15. The feedback mechanism is nonlinear and asymmetric. In exceptionally good times, the marginal positive impact of bank finance on the real economy decreases. This is because banks are virtually sellers of call options, with limited upsides (the maximum banks can make on a loan portfolio is limited by the non-contingent lending rates). In exceptionally bad times, however, the downside is practically unlimited. Sizeable adverse shocks hit the performance of loans granted not only to marginal borrowers, but also to all legacy borrowers. The resulting losses thus grow rapidly.

16. The real sector of the model is calibrated on a small open economy to broadly match the characteristics of emerging Central and Eastern European countries. Parameterizing the macro-financial linkages is less straightforward. Since responses to shocks and external scenarios are not additive in globally nonlinear models, a much larger (by several orders of magnitude) number of simulations and parameter combinations would have to be examined to achieve the same confidence as when calibrating regular business cycle models. In addition, there is little data relative to periods of nonlinear financial distress to draw upon. Thus, as discussed earlier, quantitative predictions of nonlinear models should not be taken literally.

B. Design of Simulation Experiments

17. Simulations seek to capture a sustained period of capital inflows followed by a sudden reversal. An initial period of three years of cheap foreign (nonresident) finance5 and capital inflows (Phase I, blue background in the charts) is followed by a sudden and unexpected reversal of capital flows (Phase II, white background in the charts). After the reversal, the economy gradually returns to its normal state. The simulations are designed as stress scenarios (that is, low-probability though still plausible scenarios with large adverse impact on the economy). They do not represent the most likely baseline projections or forecasts.

18. Phase I is set up as a prolonged period of low foreign financing costs. The cost of foreign financing falls by 200 bps. Note that the exact reason for such a decrease in the cost is largely irrelevant from the point of view of the domestic economy. In the real world, the drop represents the lower interest rates that followed UMP in AEs.

19. During Phase I, banks and borrowers do not internalize the risk of a reversal. All agents behave as if the low cost of foreign financing were going to continue indefinitely. This myopia gives rise to an externality; and the risk associated with bank lending is underpriced.

20. Phase II sees first a rapid increase in the cost of foreign financing, and then a gradual return to normal. The cost of foreign financing increases initially by 300 bps (thus overshooting the normal level by 100 bps), and then converges back to normal within about five years.

21. Four different scenarios are simulated to provide a full account of the nonlinearities arising in response to macrofinancial vulnerabilities. The model considers two different economies: one is a “resilient” economy because of a very limited proportion of foreign currency loans (5 percent); the other is a potentially more “vulnerable” economy with a much higher proportion of bank loans in foreign currency (50 percent). For each economy, two simulations are run: the first with a linearized version of the model, and the second using a global nonlinear solution. Conventional linearized models will somewhat overestimate the upside during good times, and greatly underestimated the downside during bad times.

22. Throughout the simulations, no pro-active policies in response to the build-up of macrofinancial risks are considered. Although the model allows for several types of countercyclical macroprudential policies (such as capital surcharges), the simulations assume that policy remains passive.

23. Finally, the share of FX loans is given parametrically, and the choice of currency is not endogenous in the model. There are various reasons why some economies experience FX lending while others rely more on local currency lending. On the demand side, the reasons relate to the credibility of monetary and other policies, the existence of risk spreads, and to myopia in assessing future exchange rate risks. On the supply side, decisions can be affected by the prudential regulation on FX lending.

C. Phase I—Capital Inflows

24. The lower cost of finance, followed by an appreciation of the domestic currency and a rise in asset prices, results in faster credit growth. Domestic households and firms use the additional purchasing power to increase their demand for local goods, imports, and assets (such as housing, productive capital, stocks). As a result, real economic activity experiences a boom. Import demand in particular is strong, owing to the exchange rate appreciation. The current account thus deteriorates, while net foreign liabilities increase.

25. The upturn in credit and real economic activity is amplified in an economy with a higher proportion of foreign exchange loans. The exchange rate appreciation further improves the borrowing capacity of households and firms (by reducing the loan-to-value or debt-to-income ratios). At the same time, the risk of future defaults in the event of a large depreciation also grows, but is not fully internalized by banks and borrowers.

26. The capital adequacy ratios of banks decline over time. Banks choose to hold thinner regulatory capital buffers, as they perceive lending to be safer as the wealth of borrowers increases. This poses the classical problem of pro-cyclical capital requirements.

27. In both economies, the linearized simulations over-predict the upturn. Nevertheless, the differences look relatively innocuous compared with those observed after a reversal, in Phase II. Note that capital adequacy ratios are not reported for the linearized simulations since credit risk, the main determinant of capital ratios, does not exhibit first-order dynamics in the model.

D. Phase II—Turning of the Cycle and Capital Outflows

28. The unanticipated increase in the cost of foreign financing, and the reversal in capital flows, reduces the sustainable level of the economy’s debt. Both types of economies (regardless of the share of foreign currency loans) must undergo current account adjustments. The current account adjustments are achieved by reductions in consumption and investment (and hence demand for imports), and by improvements in real exports facilitated by the sudden depreciation of the exchange rate.

29. Households’ and firms’ access to bank credit deteriorates rapidly, driven by the currency depreciation and fall in asset prices. At the same time, the currency depreciation lowers the real income of households, with pass-through to wages assumed to be more sluggish. These two factors undermine aggregate demand.

30. In the economy with high foreign currency lending, the impact is considerably magnified by valuation effects. The amount of outstanding bank credit expressed in local currency reaches very high levels, as the valuation effect of exchange rate depreciation outweighs an actual drop in the effective volume of bank lending.

31. Nonlinear feedback mechanisms between real and financial variables amplify the effects of the shock. The simultaneous depreciation of the currency and fall in asset prices results in sharp increases in non-performing loans. As banks write-off the unexpected losses on their loan books, capital buffers deteriorate. The effect is negligible in the case of the resilient (low FX) economy, yet large for the vulnerable (high FX) economy in which capital buffers drop by as much as 1 percentage point. Banks cut back lending in order to recapitalize. Banks also raise the price of bank credit (lending spreads), and, simultaneously tighten lending conditions (effectively rationing credit). The first is especially important to increase profit margins and thus rebuild capital. The fall in credit triggers a vicious circle and further depresses demand, depreciates the currency and undermines asset prices. In turn, these developments raise non-performing loans, further fueling the process.

Figure 3.
Figure 3.

Model-Based Simulation Experiments

Citation: Policy Papers 2013, 013; 10.5089/9781498341349.007.A001

Effects of UMP on Bond and Equity Flows and Prices6

This section provides new evidence on the effects of UMP on flows into and out of country bond and equity mutual funds, as well as the impact on bond yields, and equity prices. A range of techniques provides a relatively holistic picture of the likely effects of UMP, including: (1) factor analysis of the role of global, regional, and country factors in driving weekly flows in and out of country mutual funds; (2) regressions to see how these weekly flows relate to purchases of assets by the U.S. Federal Reserve (FED), Bank of England (BoE), European Central Bank (ECB), and Bank of Japan (BoJ) as well as UMP announcements; (3) regressions to analyze the extent to which weekly portfolio flows affect asset prices; and (4) event studies on daily asset prices to assess if forward guidance announcements had an additional impact on asset prices independent of asset purchase announcements.

The results underline that looking at announcements of UMP provides only part of the overall impact of these policies. In particular, mutual fund flows are generally found to respond more to actual UMP bond purchases than to UMP announcements. Furthermore, the pattern of flows varies by UMP program in an intuitive manner given market conditions.

  • When the Fed used mortgage backed securities (MBS) and Treasury purchases to stabilize markets the results find that money initially flowed out from global markets and then into advanced markets.

  • By contrast, the most recent program of asset purchases (U.S. Large Scale Asset Purchase 3 (LSAP3)), which has been undertaken at a time when market conditions have been more stable, initially resulted in a synchronized flow into emerging markets bonds and non-U.S. equity funds.

  • The Fed’s May 22 tapering announcement is associated with a generalized repricing of risk, inducing a notable increase in the level of correlation of flows—especially EM bond flows—as U.S. LSAP3 and Japanese LSAP flows were partly reversed.

Mutual fund flows do appear to significantly affect asset prices. The evidence is clearest with respect to equities, while similar relations seem to hold in other asset markets.

Finally, forward guidance announcements do seem to have had a separate impact from announcements of asset purchases. The surprise effect of central banks’ announcements embedding forward guidance has a strong impact on stock prices and foreign currencies.

Three general conclusions are: different forms of UMP operate through different mechanisms, with purchases of assets mattering for flows and likely asset prices in addition to announcement effects; the impact of UMP policies also varied with market conditions, with the boost to domestic markets per dollar of asset purchases likely falling as leakage to the rest of the world rose; and that the size of the global component in EM bond flows implies this market may be a particularly important potential source of global risk.

A. Common Dynamics of Bond and Equity Funds Flows across the Globe: Risk-on/Risk-off Movements and Changes in Cross-correlations

32. This section uses factor analysis to look at the role of common factors in explaining flows into and out of bond and equity market mutual funds and exchange-traded funds (ETFs). The common dynamic properties of equity and bond flows across a large of pool of advanced and emerging markets are used to assess: (i) the extent to which mutual funds flows have been driven by global risk-on/risk-off movements in financial markets; (ii) whether the share of volatility in funds flows associated to such global factors has changed over time; and (iii) whether—time-wise—there is any relation between shifts in the share of volatility in bond and equity funds flows due to global factors and the implementation of UMP.

33. A Bayesian dynamic latent factor model was used to estimate common dynamic components in two different kinds of portfolio flows (bonds and equities) in our 42-country sample which have been divided in nine groups of countries having similar characteristics (called “regions”).7 In this way, it simultaneously estimates (i) a dynamic factor common to all aggregates, regions, and countries (the global factor); (ii) a set of nine regional dynamic factors common across aggregates within such a region; (iii) 42 country factors to capture dynamic comovements across the net flows of the two asset markets within each country; and (iv) a component for each asset market that captures idiosyncratic dynamics. By design, the dynamic factors capture all intertemporal cross-correlation among the observable variables.8

34. The study relies on the EPFR Global dataset. The EPFR Global database contains weekly portfolio investment (net) flows by more than 14,000 (mutual and ETF) equity funds and more than 7,000 (mutual and ETF) bond funds, with US$8 trillion of capital under management. Although this represents only less than 20 percent of the market capitalization in equity and in bonds for most countries, generally with a lower proportion for bonds compared to equities, EPFR data can be deemed as a fairly representative sample of global flows, closely matching portfolio flows stemming from BOP data (Jotikasthira and others, 2012). More details on the features of the dataset are also provided in Fratzscher and others (2013). A key strength of the data is the high (weekly) frequency of reported flows and its broad geographic coverage, for both AEs and EMEs. In the study, we use data for 25 AEs and 17 EMEs, over the period from January 1, 2007 to August 26, 2013. The EPFR data by country were adjusted for the increasing coverage of funds over time which, in aggregate, almost halves the amount observed bond and equity flows by the end of the sample compared to the raw data, with very similar rate of shrinkage for advanced and emerging markets.

35. Common dynamics of bond and equity funds flows across a large pool of advanced and emerging markets point to synchronized outflows following the Fed’s tapering announcement on May 22, 2013. Figure 4 presents the mean of the posterior distribution of the global component in equity and bond flows. Fluctuations in this global factor are likely to reflect global risk-on/risk-off movements in financial markets and global liquidity conditions, with persistent and synchronized outflows from the beginning of the crisis (August 2007) until mid-2009; significant synchronization of outflows between August 2011 to February 2012—at the peak of the euro area debt crisis; and, strikingly, a record large global outflow following the Fed’s tapering announcement on May 22, 2013, possibly reflecting increased international participation in local bond markets over time.

Figure 4.
Figure 4.

Bond and Equity Flows—Global Factor

Citation: Policy Papers 2013, 013; 10.5089/9781498341349.007.A001

Sources: EPFR dataset and staff calculation.

36. More generally, the sources of volatility in bonds and equity funds flows have been varying over time, across asset classes, and across countries.9 Figure 5 reports the proportion of weekly portfolio flows explained by the global factor for three groups of countries: emerging markets (red line); the four major users of UMP (the U.S., euro area, Japan, and the U.K., green line); and non-UMP advanced markets (black line). Equity markets showed much more coherence than bond markets in the early days of the crisis, with the global factor explaining 30–80 percent of equity flows versus 10–20 percent of bond flows. This pattern switched abruptly with the launch of the Fed’s QE1, in November 2008. Since then, the global factor has explained a surprisingly consistent 80 percent of the variation in EMEs bond flows. The proportion for advanced economies is lower and varies more with the ebb and flow of the crisis—rising, for example, in the summer of 2011 as euro area concerns increased. A similar ebb and flow, although with somewhat different triggers, is true of equity flows after QE1, with the global factor explaining more of EME equity flows than those of advanced countries. In short, emerging market bonds were at the center of “risk-on/risk-off” behavior.

Figure 5.
Figure 5.

The Role of Push Factors in Explaining Flows’ Volatility Changes over Time

Citation: Policy Papers 2013, 013; 10.5089/9781498341349.007.A001

Sources: EPFR dataset and staff calculation.

37. Since May 22, the role of global factors in explaining variations of portfolio flows seems to have jumped, pointing to a generalized repricing of risk (Figure 5). This suggests that the increase in market volatility following the recent Fed’s tapering announcements are less due to idiosyncratic EME weakness, but rather are primarily driven by push factors which are common to EMEs and non-UMP AEs alike. Table 2 provides a more granular idea of the portfolio rebalancing occurred over the period May 22 to August 26 and, most importantly, the share of the actual flows which is due to the estimated global factor—virtually 100 percent for bond flows. Overall, following the tapering announcement, we saw flows out of EMEs and AEs bond funds and primarily into U.S. equity funds. For EMEs bond funds, we estimate outflows for US$10 billion (approximately 0.2 percent of EMEs bond market size). For AEs bond funds, we estimate outflows for US$11 billion (approx. 0.1 percent of AE bond market size), with US$7.6 billion representing U.S. bond funds alone. For equity funds, we saw outflows from EMEs for circa US$20 billion, US$16 billion of which flowing into U.S. equity funds and the rest into Japanese and European equity funds.

Table 2.

Impact of the Global Factor on Bond and Equity Funds Flows over May 22 to August 26

(US$ millions unless otherwise indicated)

article image
Source: EPFR database; IMF 2013 April GFSR; and staff calculations.

38. While the ratios to market size are low, it should be recalled that mutual fund flows represent only less than 20 percent of the overall portfolio flows into the economy. This suggests a large switch in portfolios, particularly for bond markets where the EPF data covers a smaller proportion of overall flows.

B. The Role of UMP Announcements and Actual Central Banks’ Asset Purchases in Driving Bond and Equity Funds Flows

39. To date, the main approach to looking at the impact of UMP policies has been to use event studies to examine the impact of announcements on U.S. and foreign asset prices at high frequencies (e.g., daily or less). While useful, such an approach has inevitable limitations. In particular, it can only look at the effect of initial announcements, which is particularly unfortunate for a policy that partly works through actual purchases of bonds. Following Fratzscher and others (2013), this section uses a similar econometric approach to event studies—looking at the synchronization between UMP policies and flows into bond and equity fund flows that week or the following one—but broadens the analysis to look at the impact of announcements and of actual purchases.

40.Weekly data ending on Wednesdays were collected for flows into bond and equity mutual funds from the EPFR data set and on asset purchases from S4 central banks’ balance sheets. The EPFR data by country were adjusted for the increasing coverage of funds over time (as discussed in section A). Daily data on other asset price variables were also converted to weekly changes ending on Wednesday (e.g., commodity prices, VIX, etc.).

41. Given these data, the sample was divided into periods covering distinct central banks’ operations. These were the Fed’s announcements and purchases for LSAP1A (MBS purchases), LSAP1B (Treasury purchases), LSAP2, Operation Twist, LSAP3, and tapering speech on May 22, 2013; BoE’s announcements and asset purchases for LSAP1, LSAP2 and FLS; ECB’s announcements on outright monetary transactions (OMT) and ‘whatever it takes” speech, as well as the conventional rate cut on May 2, 2013; ECB’s actual long term refinancing operation (LTRO) liquidity provision and securities market program (SMP) purchases; BOJ’s announcements and asset purchases for LSAP and QQME as well as pre-LSAP asset purchases. More information on the timing of these announcements and programs is given in Appendix Table 2 (reproduced from IMF 2013a).

Appendix Table 2.

Selected Recent Unconventional Monetary Policies

article image
article image
article image
Source: Country authorities.Note: * Event is included in multiple entries.1/ Announcing an inflation target is more than just committing to temporarily loose policy in the future, instead it provides information about a permanent change in monetary policy.

Dates include 3/14/2011, 8/4/2011, 10/27/2011, 2/14/2012, 4/27/2012, 7/12/2012, 9/19/2012, 10/30/2012, and 12/20/2012, 1/22/2013.

42. Adjusted EPFR flows seem to be a reasonable proxy for gross international bond to emerging markets and equity flows to all economies except but large money centers. Table 3 reports the correlation of EPFR weekly data on flows into bond and into equity funds aggregated into quarters with the corresponding flows from the balance of payments since 2008Q1 as well as a measure of the proportion of balance of payment flows represented.10 For emerging markets the correlations for both bond and equity markets are generally around 0.5, suggesting a reasonable correspondence between the two series. The correlations are also generally fairly high for advanced market equity flows, although this is not true for money centers such as the U.S., U.K. and Switzerland.

Table 3.

Correlation and Relative Size of Adjusted EPFR Data and BoP Data

(2008–12)

article image
Sources: EPFR database; IMF BOPS; and staff calculations.Notes: Shaded values use the average of the region as BoP data were unavailable.

43. With only a few exceptions, however, the correlations are low for AE bond flows. In interpreting these results several factors should be considered. First, most AEs have deeper financial markets where domestic bonds are often used as collateral, which may distort the balance of payment data for bond flows. In addition, recall that these data reflect all flows to bond and equity funds, regardless of the domicile of the investor. Hence, the data can reflect the behavior of domestic investors. Certainly, this seems to be true for the U.S. where EPFR flows represent an implausibly large 64 percent of foreign equity inflows and 19 percent of foreign bond inflows. As will be discussed below, these flows seem to affect corresponding asset prices in all economies, suggesting that they reflect useful information about market behavior. Overall, we conclude that these data reflect a reasonable proxy of domestic market conditions in bond and equity markets.

44. The basic empirical approach relates weekly flows into bond and equity mutual funds with UMP policy announcements, central banks’ asset purchase programs, and other conditioning variables.11 More specifically:

Flowit=αi+βiAnnouncet+γiPurchase +ηiconditioning variablest+εit(1)

where Flowit is the flow of funds into bond/equity mutual funds for country i at time t, αi is a country-specific constant term, Announcet are 0/1 dummy variables for weeks with major UMP announcements, Purchaset is the amount of securities purchased by a central bank in week t, and εit, is an error term.12

45. In addition to this baseline specification, two other variations were estimated that test for simultaneity bias. The first variation uses lags all of the right hand side variables to ensure that purchases are not being affected by contemporaneous flows into bond and equity funds. This specification produced extremely similar results to the base case, suggesting that reverse causality is not a major issue. The second specification replaces actual weekly purchases each week with dummies that are one during the period each purchase program is active and zero at other times, hence eliminating any feedback between market conditions and the exact sums bought in programs.13 This specification gave similar results in terms of coefficient significance, albeit with some differences in terms of the size of implied flows.

46. While reverse causality does not appear to be a major issue in the regressions, the results do seem to reflect the impact of overall market conditions as well as UMP policies. While the conditioning variables—the VIX, oil prices, and non-oil prices—help to explain some of the volatility in EPFR fund flows, it is equally clear that the fact the Fed purchases of asset backed securities in the U.S. LSAP1A program were associated with strong outflows for bonds funds largely reflects market turmoil. Two observations are relevant here. First, the associations documented in these regressions remain important in charting the path of UMP. Second, bond and equity fund flows are more likely to be directly associated with programs initiated during periods of relative market calm, such as U.S. LSAP3.

47. The U.K., euro area, and Japan programs were excluded from some regions where their effects were assumed to be small. In some cases, results for these programs gave implausibly large estimated flows given the limited links between the source country and the recipient, likely reflecting the impact of market conditions discussed above. This was particularly true for fund flows to the U.S. and Canada. In addition, Latin American emerging market regressions excluded U.K. and Japan programs, while Japan programs were also excluded from European countries.

48. Tables 4 and 5 report the coefficient estimates for the baseline specification. Shaded coefficients are significant at the one percent level (dark blue), five percent levels (medium blue), and ten percent level (light blue). Insignificant coefficients are marked in gray. The results reveal some interesting differences in the impact of different programs. For bond flows, for U.S. LSAP1A, U.S. tapering, and Japan LSAP both the announcement and actual purchases had large numbers of significant coefficients, while purchases seems to have been the driving force for U.S. LSAP1B, U.S. LSAP2, U.S. LAP3, U.K. FLS and euro area LTRO liquidity programs. Other initiatives attract smaller numbers of significant coefficients. In addition, as might be expected since UMP asset purchases mainly involved bonds, the number of significant coefficients is generally higher for bond flows than equity flows.

Table 4.

Regression Results for EPFR Bond Fund Flows

article image
article image
Sources: EPFR database; IMF 2013 April GFSR; and staff calculations.Notes. Dark, medium, and light blue ndicate the coefficient is signifinat at the 1, 5, and 10 percent level. Gray indicates an insignificant coefficient.