The Stategic Petroleum Reserve and Oil Prices

Transcription

1 The Stategic Petroleum Reserve and Reid Stevens UC Berkeley October 8, / 52

2 Question: Does the SPR affect crude oil prices? Answer: Yes, but not as intended. Assumption Data Crude Oil Release Oil Price Oil Price Crude Oil Purchase Oil Price Oil Price 2 / 52

3 (SPR Purchases) (SPR Releases) 3 / 52

4 : What is the SPR? 730 million barrels of crude oil Response to Arab Oil Embargo To reduce the adverse economic impact of a major petroleum supply disruption Controlled by the President SPR Storage Locations 4 / 52

5 : How big is the SPR? 67% of total domestic crude stocks The equivalent of 148 days of net imports Cost $100 billion to date Commercial Stocks 33% 67% Government-Controlled Stocks Total Crude Oil Stocks: 1.1 billion barrels 5 / 52

6 : When is it used? SPR Releases International supply disruptions Domestic supply disruptions Non-emergency 8% 4% 0% -4% -8% -12% -16% SPR Purchases/Releases as a Share of US Oil Production 6 / 52

7 : Why would we care? Every recession (with one exception) was preceded by an increase in oil prices, and every oil market disruption (with one exception) was followed by an economic recession. Hamilton (2011) If the SPR can decrease oil prices, it would be very valuable 7 / 52

8 : Why would we care? Over half of global crude oil stocks in government-controlled reserves 26 IEA and 27 EU member countries must hold strategic reserves Russia, China, Japan also have large strategic reserves Global Strategic Reserves 8 / 52

9 : Literature Answers the question: How should the SPR be managed? 1980s US SPR: Teisberg (1981); Chao and Manne (1982) 2000s China s SPR: Wei et al. (2008); Han et al. (2014) Little empirical work on the effect of the SPR on Oil prices Considine (2006) 9 / 52

10 : Literature Why don t we know the SPR price effect? Spurious correlation Causal direction unclear Mutual dependence on a common cause process difficult to model 10 / 52

11 : Literature Killian (2009) monthly global oil market model: where A 0 Y t = α + 24 i=1 A iy t i + ε t Oil Supply t Y t = Oil Demand t Oil Price t 11 / 52

12 : Add SPR variables to weekly US oil market model: where A 0 Y t = α + 40 i=1 A iy t i + ε t Oil Supply t Oil Demand t Y t = SPR Purchase t SPR Release t Oil Price t 12 / 52

13 : In this model, A 0 Y t = α + 40 i=1 A iy t i + ε t The causal effect of SPR policy on crude oil prices is: P rice t+h, h = 1, 2, 3,... ε SP t R 13 / 52

14 : But we only estimate the reduced form: where Y t = β + 40 i=1 B iy t i + e t e t = A 1 0 ε t Useful forecasts but no causal interpretation 14 / 52

15 : Identifying Assumptions To identify the causal effect, assume a temporal ordering of variables (exclusion restriction). This is equivalent to restricting A 1 0 in e t = A 1 0 ε t 15 / 52

16 : Identifying Assumptions Exclusion Restriction p t = α 1 q t + α 1 q t α k p t q t = β 1 p t + β 1 p t β k q t / 52

17 : Identifying Assumptions Exclusion Restriction p t = α 1 q t + α 1 q t α k p t q t = β 1 p t + β 1 p t β k q t / 52

18 : Identifying Assumptions Partition Y t into slow-moving, policy, and fast-moving variables: Slow-moving: Oil supply & oil demand Variables: SPR purchases & SPR releases Fast-moving Variable: Oil Price 18 / 52

19 : Identifying Assumptions Assumption 1: Oil supply shocks are exogenous and do not respond contemporaneously to other structural shocks Motivation Production schedule changes costly Import transit time > 1 week 19 / 52

20 : Identifying Assumptions Assumption 2: Oil demand immediately responds to oil supply shocks, but not to SPR policy or oil price shocks Motivation Economic production changes costly 20 / 52

21 : Identifying Assumptions Assumption 3: SPR policy can respond immediately to oil supply and demand shocks, but does not respond immediately to oil price shocks Motivation Meeting-filled policy process 21 / 52

22 : Identifying Assumptions Assumption 4: Oil prices can respond immediately to all other structural shocks Motivation Prices move quickly 22 / 52

23 : Data Weekly Data: June 10, October 3, 2014 Oil Supply: Domestic Crude Production and Imports Oil Demand: ADS Business Conditions Index SPR : Published by DOE Oil Price: West Texas Intermediate Spot Price (Real) 23 / 52

24 : Results SPR Impulse Response Functions 24 / 52

25 : Results Release Purchase Average Structural Residuals for Subperiods 25 / 52

26 : Criticisms Criticisms (Rudebusch, 1998) Timing assumptions not credible (simultaneniety) reaction functions are naive (omitted variables) Estimated structural shocks do not always match futures markets shocks 26 / 52

27 Natural experiment or Narrative approach Romer and Romer (1989, 2004); Ramey and Shapiro (1998); Stock and Watson (2010) Identify the effect of a policy (without exclusion restrictions) with an instrument that is Correlated with unexpected policy changes Uncorrelated with all other structural shocks 27 / 52

28 : Data I construct the SPR purchase schedule as an instrument, Z t, for actual purchases. Purchase schedules set months before actual purchases and not publicly announced E(Z t ε i t) = 0, i P urchase DOE reluctant to give purchase schedule exemptions E(Z t u P urchase t ) 0 28 / 52

29 : Data Barrels per week 160, ,000 80,000 40, SPR Purchase Schedule (Red) and Actual Deliveries (Blue) 29 / 52

30 : Partition Y t into the SPR purchase variable (P 1 1) and all other variables (X 4 1 ). e X t ε x t A 11 A 12 e t =, ε t = A 0 = A 21 A 22 e P t Now, A 0 e t = ε t becomes: ε P t A 11 e X t + A 12 e P t = ε X t A 21 e X t + A 22 e P t = ε P t 30 / 52

31 : Which simplifies to, To estimate ε P t e X t e P t = A 1 11 A 12e P t + A 1 11 εx t (1) = A 1 22 A 21e X t + A 1 22 εp t (2) 1 Use Z t as instrument for e P t in (1) and estimate A 1 11 A 12 2 Estimate ε X t = A 1 11 εx t, as e X t + A 1 11 A 12e P t 3 Use ε X t as an instrument for e X t in (2) to estimate A 1 22 A 21 4 Finally, estimate ε P t = e P t + A 1 5 ε P t give us ε P t to scale, Oil P ricet ε P t+h 22 A 21e X t, h = 1, 2, / 52

32 : Results Impulse Response Function (Purchase ) 32 / 52

33 Cochrane and Piazzesi (2002); Bernanke and Kuttner (2005) Directly estimate structural policy shock (ε release t ) from daily futures market data If you are right, you are done Oil Price t+h, ε SPR h = 1, 2, 3,... t 33 / 52

34 Assumptions: The change in crude futures price (P t P t 1 ) captures the unexpected component of the SPR announcement The change in crude futures price is driven only by SPR announcement 34 / 52

35 Examples 35 / 52

36 : Results Impulse Response Function (Release ) 36 / 52

37 : Problematic assumption: The change in crude futures price is driven only by SPR announcement To make this assumption more plausible, use an alternate financial instrument WTI-Canadian crude oil futures spread 37 / 52

38 : Results Impulse Response Function (Release ) 38 / 52

39 Question: Does the SPR affect crude oil prices? Answer: Yes, but not as intended. Assumption Data Crude Oil Release Oil Price Oil Price Crude Oil Purchase Oil Price Oil Price 39 / 52

40 interacted (Towbin and Weber, 2013) Add uncertainty (V t ) and uncertainty-spr policy interaction terms to the price equation where, 40 Y t = B 0 + (B i Y t i + C i Y t i V t ) + DV t + e t i=1 V t = WTI Spot Price Volatility (90-day) 40 / 52

41 : Data Oil Market Index 41 / 52

42 Use the V high = 90th percentile volatility to estimate, where, Y high t = F high + 40 i=1 Ĝ i high Yt i + ê t F high = B 0 + DV high high Ĝ i = Bi + ĈiV high 42 / 52

43 Use the V low = 10th percentile volatility to estimate, where, Y low t = F low + 40 i=1 Ĝ i low Yt i + ê t F low = B 0 + DV low low Ĝ i = Bi + ĈiV low 43 / 52

44 : Results Low Volatility High Volatility SPR Purchase Impulse Response Functions 44 / 52

45 : Results Low Volatility High Volatility SPR Release Impulse Response Functions 45 / 52

46 mechanisms have been proposed in the literature (Considine, 2006): OPEC reduces production in response to SPR releases Gulf Coast commercial stocks absorb SPR releases 46 / 52

47 To find the effect of SPR policy on OPEC production and Gulf Coast crude oil stocks: First, convert weekly structural shocks to monthly shocks, ε m t = 1 4 ε it 4 i=1 Then, regress OPEC production and regional stocks on monthly shocks, OP EC t = α + Stocks t = α + 9 i=0 9 i=0 ε m t ε m t + u t + u t 47 / 52

48 SPR-OPEC Impulse Response Functions 48 / 52

49 SPR-Gulf Coast Crude Stocks Impulse Response Functions 49 / 52

50 : Do not purchase oil for the SPR when oil market uncertainty is high If you want to lower oil prices, try another policy 50 / 52

51 : Coordinate strategic reserve purchases across countries SPR-OECD Strategic Reserves Impulse Response Functions 51 / 52

52 Thank You 52 / 52