The Effects of Quantitative Easing on Interest Rates (KVJ)

Transcription

1 The Effects of Quantitative Easing on Interest Rates (KVJ) Minjoon Lee December 6, 2011

2 Outline 1 Motivation 2 KVJ(2010) Introduction Simple version of model, prediction and evidence Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Implications 3 KVJ(2011) Introduction Channels Evidence for QE1 ( ) Evidence for QE2 ( ) 4 Conclusion

3 Zero Lower-bound ST interest rate is already at zero lower-bound. But we still have a room to lower down LT rates. Hamilton and Wu (2011), The Effectiveness of Alternative Monetary Policy Tools in a Zero Lower Bound Environment", JMCB

4 Which interest rates we aim at? The most relevant rates are those on long-term, illiquid and not that safe bonds. Compared to the Treasury bonds, these lack safety and liquidity. 1 Pricing of Liquidity and Safety: KVJ (2010), The Aggregate Demand for Treasury Debt" 2 How QEs affect pricing of those features: KVJ (2011), The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy"

5 Introduction There is something in Treasury Securities Which properties are in Treasuries but not in Aaa? : Safety and Liquidity The pricing on these features can be understood as seignorage

6 Introduction What the paper does 1 Identification of liquidity and safety pricing separately : compare yields between assets with the same characteristic except for the dimension of liquidity or safety : see how the pricing is changed with the change in the supply of Treasury securities 2 Show that Bank issued money is subsititute for Treasury securities : this again shows that the Treasury securities share some of the characteristic of Money

7 Introduction The strengths of the approach By using spreads in estimations (diff-in-diff), 1 Estimates can be free from the omitted variable problem 2 Estimates can be free from the endogeneity issues

8 Simple version of model, prediction and evidence Simple version of the Model Investor s problem Max E t=1 βt u(c t ) where C t = c t + ν(θt A, GDP t; ξ t ) θt A = θt T + k P θt P : holding of convenience asset ξ t : Preference shock Here ν( ) function is a reduced form expression of the convenience coming from Treasuries (and others). Hence this is not much different from the M-I-U model. We assume ν( ) is homogeneous of degree 1 in θ A t, GDP t. Hence, we define υ( θa t GDP t ; ξ t )GDP t ν(θ A t, GDP t; ξ t ) Using this preference and the methods of CAPM, we can derive equilibrium price and yields of securities.

9 Simple version of model, prediction and evidence Prediction 1: Short-term spreads Consider 1-year maturity Treasury (T) and corporate bond (C) which does not have convenience. P T t+1 = PC t+1 = 1 Short-term spreads S t,1 i C t i T t = υ ( θt t +k P θ P t GDP t ; ξ t ) + λ t E t [L t+1 ] + Cov(M t+1, L t+1 )/E t [M t+1 ] First term: The price on the convenience feature Second term: Expected loss of default Third term: The evaluation of the risk using the kernel Prediction 1 Assuming that υ () < 0 and θp t θt T spreads decrease. > 1 k P, when θ T t increases the

10 Simple version of model, prediction and evidence Prediction 2: Long-term spreads Now we consider τ-year maturity case. Then by following a similar path, Long-term spreads S t,τ = t+τ 1 t+τ 1 j=t j=t 1 τ E t[υ (θj A /GDP t ; ξ t )] + t+τ 1 1 j=t τ E t[λ j D j ] 1 τ cov t(m j+1, R j+1 ) The way of interpretation is the same as before. Prediction 2 Under the same assumptions as before, when the supply of Treasuries go up, the long-term spreads go down.

11 Simple version of model, prediction and evidence Evidence 1 and 2 Here, Debt term is measured by the total market value of Treasuries (held by public). Considering the sensitivity of the market value w.r.t. the yields, is it appropriate to use market values?

12 Simple version of model, prediction and evidence Quantifying the effects Aaa-Treasury (long-rate) Coefficient: Implies that a decrease of one s.d. drop of Treasury-Debt ratio from its mean (0.426 to 0.233) increases the spread by 45 bps. The effect on Baa-Treasury spread is much greater. CP (with A1/P1) - Bills (short-rate) Coefficient: Hence the implied effect is very similar to Aaa-Treasury s case.

13 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Model separating liquidity and safety Now we conceptually distinguish convenience coming from liquidity and safety. (Technically there is nothing new.) υ T,j ( ) = υ liq ( θt t +k liq θ P,liq t GDP t ; ξ liq for j = short, long t ) + υ j safe ( θt,j t +k j safe θ P,j safe t GDP t ; ξ j safe t ) Why we do this? To cancel out one part when we compare assets with difference in only one dimension.

14 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Justification for the safety term The expected default loss is already captured in usual CAPM valuation. Then why we need to separately include safety term into the convenience function? For extremely safe assets, there is more than we can explain by the difference in default rates.

15 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Justification for the safety term Why is it maturity specific? Virtue of having fixed stream of income for intended periods. (Consider pension funds) This is why Agency MBS s are considered not to have the same safety attributes as Treasuries.

16 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Prediction 1: Pricing of short-term safety Which assets to compare? Commercial papers, rated P1 and P2 (both are similarly illiquid) Prediction 1 If the supply of short-term Treasuries increases, P2-P1 spread (the price of short-term safety) will shrink. Note that in running regressions we instrument the supply of short-term Treasuries by that of Total Treasuries, since the maturity structure may be endogenous to observed rate-differences. (Consider QEs)

17 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Prediction 2: Pricing of long-term safety Which assets to compare? Long-term corporate bonds, rated Aaa and Baa. (Both are illiquid. Also Aaa is very safe: see previous picture) Prediction 2 If the supply of Long-term Treasuries increases, Baa-Aaa spread (the price of long-term safety) will shrink. Again θ T,long t is instrumented by θ T t.

18 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Prediction 3: Pricing of Liquidity Which assets to compare? One year Treasury and FDIC insured bank deposit (Both are very safe, but the latter is much less liquid) Prediction 3 If the supply of Treasuries increases, FDIC insured deposit - 1 year Treasury spread (the price of liquidity) will shrink.

19 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Evidence on Predictions

20 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Quantifying the Effects Baa-Aaa (long-term safety) Coefficient: One s.d drop of long-term Treasury supply from its mean (from to 0.021) increases the spread by 41bps. P2-P1 (short-term safety) Coefficient: One s.d drop of short-term Treasury supply from its mean (from to 0.104) increases the spread by 26bps. FDIC insured deposit-bills (liquidity) Coefficient: One s.d. deviation drop of Treasury supply increases the spread by 61 bps.

21 Separating pricing on Liquidity and Safety: Model, Prediction and Evidence Prediction 4: Impact of Treasury supply on supply of substitute assets Last check to see whether Treasury security shares some attributes with money is to see Bank-issued money and Treasuries do (partially) substitute each other. Prediction 4 If supply of money is price elastic, then θ Money t and θ T t will be negatively correlated. In this case, M1 is excluded. Why? 1 M1 also has medium of exchange attribute. This may make comparison complicated. 2 Reserve supply is governed by Fed. (So maybe determined not just by price.)

22 Implications Implication 1: Seignorage Borrowing with lower interest rate: seignorage Here they estimate the piecewise linear convenience function, b 1 max[b 2 Debt/GDP, 0] They found out that the average seignorage in the history ( ) has been 72 bps (Baa-Treasury), where 46 bps among this comes from liquidity (Aaa-Treasury). One interesting thing: b 2 is estimated to be Means?

23 Implications Implication 2: Money demand + 3: Equity premium puzzle 1 Money demand In estimating money demand function, if we do not include Treasuries (with appropriate weightings) we will end up with a very unstable one. 2 Equity premium puzzle Considering the convenience yield, the spread between equity - Treasury should be larger than that suggested by usual CAPM calculation. This partly explains the Equity premium puzzle.

24 Introduction Quantitative Easing QE? Fed s purchase of long-term securities including Treasuries, Agency bonds, and Agency MBS In QE2, the purchase has been concentrated on Treasuries.

25 Introduction Goal and Methodology Goals 1 Identify different channels through which QEs affect interest rates. 2 Figure out how and why the size of effects are different on different interest rates. Methodology 1 Using daily data, identify channels which are in effect around the key announcement dates. 2 Using intra-day data, support the idea that the changes in those days reflect the effects from announcements.

26 Introduction Why identifying channels are important? Ultimately, we want to affect corporate yields and MBS rates (long-term, illiquid asset rates). Then we need to know (i) what is the effect of each component of QE and (ii) through which mechanism it propagates to those rates.

27 Channels 1. Signaling Channel 1 Mechanism Unconventionally expansionary monetary policy signals lower interest rates in the future. Fed s holding of long-term securities can be considered as commitment for low interest rates in the future (Clouse et al. (2000)) 2 Expected effects Lower interest rates for all bond markets ( Baseline interest rate goes down, not affecting pricing of prices and safety). Changes in intermediate rates are largest (at some point interest rates should come back to the normal level). 3 How we examine Price of federal funds futures contract shows the expected future federal funds rate. If there is any change in expectation through this mechanism, it will be reflected in the change in this price.

28 Channels 2. Duration Risk Channel 1 Mechanism QE Reduced duration in the market Less duration risk This will reduce the market price of duration risk (Vayanos and Vila (2009) - Preferred habitat demand) 2 Expected effects QE decreases long-term nominal interest rates. The effect will be larger when the maturity is longer. 3 How we examine Compare changes in CDS-adjusted corporate bond rates of long- and intermediate-maturities Check whether there is pattern captured by this channel

29 Channels 3. Liquidity Channel 1 Mechanism Reserve is more liquid than Treasury bonds Then QE will reduce the price of liquidity 2 Expected effects Relative yield on liquid assets goes up. 3 How we examine Check Agency bond - Treasury spread

30 Channels 4. Safety Channel 1 Mechanism Supply of long-term safety goes down Price of this goes up 2 Expected effects Relative yields of long-term safety assets go down. 3 How we examine Check CDS-adjusted Aaa-Baa spreads (note that this gives the lower bound of the effect)

31 Channels 5. Prepayment Risk Channel 1 Mechanism Prepayment risk is peculiar to MBS If QE contains purchase of MBS, this will reduce prepayment risk remaining in the market 2 Expected effects If QE contains purchase of MBS, relative yield on MBS goes down 3 How we examine If the change in MBS rate cannot be fully explained by signaling effect, it is likely that this channel is in effect.

32 Channels 6. Default Risk Channel 1 Mechanism If QE is expected to succeed in stimulating the economy, then expected default risk goes down 2 Expected effects Risk premium goes down 3 How we examine CDS (credit default swap) rate is a good measure of expected default risk. Check changes in this rate.

33 Channels 7. Inflation Channel 1 Mechanism QE is expansionary π e It can either increase (too much expansionary) or decrease (fight against deflation risk) uncertainty on inflation rate 2 Expected effects π e, Var(π e ) 3 How we examine Check the inflation swap rate, which is a good measure of market expectation on future inflation Check the swaption (an option to enter into an interest rate swap) rate, which is a good measure of market opinion on the volatility of future interest rate

34 Channels Resulting long-term rate for unsafe, illiquid assets r risky,illiq,long term = E[i safe,liq,short term ] π e +Duration P DurationsRisk + Illiquidity P Liquidity + LackofSafety P Safety + DefaultRisk P DefaultRisk + (PrepaymentRisk P PrepaymentRisk )

35 Evidence for QE1 ( ) How to approach Identify 5 dates when big announcement regarding Fed s intention on additional purchases have been made. Use OLS, where LHS is the variables (usually spreads) that we are interested in RHS contains dummies for announcement dates.

36 Evidence for QE1 ( ) Evidence 1: Signaling One problem: the maximum maturity of this contract is 2 years, so we cannot see what happens after that. We can consider that 40bps fall is upper bound for longer maturity.

37 Evidence for QE1 ( ) Evidence 2: Duration risk We cannot find clear pattern implying Duration risk channel.

38 Evidence for QE1 ( ) Evidence 3: Liquidity Across all the maturities, Agency bond - Treasury spread falls. Liquidity channel in effect.

39 Evidence for QE1 ( ) Evidence 4: Safety For both long and intermediate maturities, yields on Aaa bond fall more than that on Baa.

40 Evidence for QE1 ( ) Evidence 5: Prepayment risk Though Agency MBS is not as safe as Agency bonds, the falls are in the similar magnitude. This shows that the prepayment risk channel is in effect.

41 Evidence for QE1 ( ) Evidence 6: Default risk CDS rates fall and the size of change is larger for lower grade bonds. Hence default risk is channel in effect.

42 Evidence for QE1 ( ) Evidence 7: Inflation Channel Inflation swap rates increase, implying that the expected inflation rates went up (largest for 10 year maturity). Swaption rate falls, means that uncertainty about future inflation rate decreased.

43 Evidence for QE2 ( ) Comparison between QE1 and QE2 In QE1, all the channels except for duration risk channel were in effect. In QE2, as we will see, liquidity channel, credit risk channel and prepayment risk channel are ineffective.

44 Evidence for QE2 ( ) Liquidity Channel Muted Agency bond - Treasury spread is not affected. Hence Liquidity channel is ineffective. Why? Fed provided enough liquidity through QE1 so the liquidity came back to the normal level in this period.

45 Evidence for QE2 ( ) Prepayment Risk Channel Muted The comparison between this and the previous table shows that Fall in Agency MBS rate is fully explained by signaling effect. Hence no prepayment risk channel. Why? In QE2 Fed did not purchase MBS. Hence did not decrease prepayment risk in the market.

46 Evidence for QE2 ( ) Credit Risk Channel Muted CDS rates do not fall (actually they rise if we consider 2-day changes). Why? The stimulation announced in QE2 was less aggressive than what has been expected by market. Or market took QE2 as a signal that the current economy is in worse shape.

47 Conclusion To stimulate the economy successfully, we have to aim at some interest rates which are intimately related to real investment activity. To do that, we have to understand all the factors coming into the pricing of each asset and figure out ways to affect those factors directly (eg. Reducing specific risks from the market..)