1
New Evidence on the Financialization* of
Commodity Markets
Brian Henderson
Neil Pearson
Li Wang
February 2013
* “Financialization” refers to the idea that non-information-based
commodity investments by retail and institutional investors who buy
commodities for portfolio reasons (e.g., diversification) have
important impacts on commodity prices.
2
Background
• From 2003 to 2011, “financial” commodity investment
increased from $15 to $400 billion
• The periods of most rapid increases in financial
commodity investment approximately coincided with
the periods of significant increases in commodity
prices, especially the 2008 commodities boom
• Do these financial commodity investments have a
causal impact on commodities prices? Some
researchers say “yes,” e.g. Tang and Xiong (2011),
Mou (2010), Singleton (2012)
• Others find no evidence, e.g. Stoll & Whaley (2010),
Buyuksahin and Robe (2009, 2011)
3
Related Literature
Evidence of “financialization:”
• Purchases by non-commercial traders have causal impacts on
commodity futures prices or expected returns (Masters 2008,
Gilbert 2010, Singleton 2012, Hamilton and Wu 2012).
• Evidence from correlations (Silvennoinen and Thorp 2010;
Tang and Xiong 2011)
• Index rolling affects futures prices (Mou 2010).
No evidence for “financialization:”
• Position changes computed from CFTC data do not “Granger
cause” futures prices (Buyuksahin and Harris 2009)
• “Rolling” of positions does not impact commodities prices
(Stoll and Whaley 2009)
• Hedge fund positions rather than index investment explain the
recent increase in correlations between stock and commodity
returns (Buyuksahin and Robe 2011)
4
Total “Financial” Commodity Investment
5
Summary
• We study the impact of issues of commodity-linked
notes (CLN’s) on commodity prices. CLN’s are issued
by financial institutions and sold to retail investors
• The sample of CLN’s is useful because:
– Issuers’ hedge trades reflect the retail demand
– We know when the issuers hedge the issues, and thus know
when to look for the price impact of the retail demand
– The CLN issues do not convey information to the market,
because neither the CLN purchases nor the issuers’ hedging
trades are based on information about commodity prices
• The hedge trades executed by the financial institutions
that issue the CLN’s have significant non-transitory
impacts on commodity futures prices, consistent with
the “financialization” of commodity prices
6
What are CLN’s?
Medium-term notes issued by financial intermediaries
Payoffs based on the price (return) of a commodity, a
commodity futures contract, a commodities index, or
basket of commodities.
Example: Accelerated Return Notes (ARN)
Pricing Date: 16 Dec. 2010
Linked to price of frontmonth crude oil futures
Total proceeds: $11,316,090
No periodic interest payment
Not listed on any exchange
CLN Issuer Hedges Using Futures
Hedging
Retail Investors
purchasing CLNs
Financial
Intermediary
Futures Market
Hedging
Re-hedging
• OTC Swap
CLNs are priced based on the closing price of
underlying commodity futures on the pricing date.
On the pricing date, issuer hedges by buying futures
or swaps, which are then re-hedged using futures
8
Identification
• Trades that hedge CLN’s sold to retail investors are
plausibly exogenous
• Issuers will execute hedge trades regardless of changes
in futures prices on the pricing datethis eliminates the
“reverse causality” from price changes on the pricing
date to the hedge trades
• Sophisticated investors with valuable information are
unlikely to buy high-cost and low-leverage CLN’s rather
than low-cost futures contractsthis rules out
“common causality” in which both CLN issuances and
price movements are caused by retail investors’ private
information about commodity prices.
9
Data
• SEC/EDGAR database:
–
–
–
–
–
CLN pricing supplements (form 424B2 or 424B3)
All public issues from Jan. 2003 to Aug. 2011
Issued by 20 banks and financial intermediaries
1491 issues, 106 of which are ETNs
For each CLN: pricing date, maturity date, underlying asset,
commissions, total proceeds, listing information, CUSIP
– Total proceeds are around $59 billion, 11% of the total
“financial” commodity investments.
• Bloomberg:
– Daily closing price and open interest for futures contracts, S&P
500, Morgan Stanley Emerging Market Index, JP Morgan Bond
Index, US Dollar Index
• Federal Reserve:
– Inflation compensation (Gürkaynak, Sack, and Wright 2010)
10
Numbers of Issues and Proceeds ($millions), by Year
Agriculture
Year
2003
2004
2005
2006
2007
2008
2009
2010
2011
Total
Average
Proceeds
No.
0
0
10
12
17
25
28
9
31
132
$
398
44
1,112
3,183
957
310
923
6,926
52
Energy
No.
0
1
0
15
37
34
37
39
45
208
15
1,113
2,463
54
286
580
4,510
Platinum
and
Palladium
No.
$
0
0
0
0
0
3
225
1
50
7
192
22
231
33
697
54
21
Industrial
Metals
$
73
603
865
8,999
9,270
480
1,014
21,305
102
No.
0
0
0
1
12
13
16
16
25
83
$
26
21
7
153
131
2,428
1,214
1,670
493
6,142
Divers. Index
and Comm.
Baskets
No.
$
4
284
14
508
31
627
53
3,221
112
3,606
186
5,743
153
1,702
161
1,813
129
1,853
843
19,357
32
23
Gold and
Silver
No.
1
2
2
10
7
20
51
66
33
192
$
Full Sample
No.
5
17
43
91
185
281
286
298
285
1491
Numbers of issues and proceeds for 2011 are based on data only through
August 2011
$
310
602
1,032
4,036
6,827
23,041
13,246
4,750
5,093
58,937
40
11
Hypotheses
• Expect price impact (positive abnormal returns) around
the pricing dates of CLN issues due to hedge trades
– Due to the possibility of delayed hedge trades, we focus on
two-day (pricing date and next day) abnormal returns.
• Price impact should be increasing in the size of the
hedge trade
– Use issue size as a proxy for hedge trade size.
– Expect to see a larger price impact for larger issues.
• Do not expect significant price impact for issues linked
to diversified commodity indices because the hedge
trades are spread across the many index commodities
12
Abnormal Futures Returns
• Following Tang and Xiong (2011), we use the
“return” Rt = ln(Ft) – ln(Ft−1) on the front-month
futures contract
• We focus on the abnormal return Rt − Rt , where the
benchmark return is based on a factor model
Rt = αˆ + βˆS & P 500 RS & P 500,t + βˆEMA REMA,t + βˆUSD RUSD ,t
+ βˆBOND RBOND ,t + γˆRt −1 + γˆEMA REMA,t +1 + βˆinflation Rinflation ,t
• The factor model is estimated separately for each
combination of commodity and pricing date using
data from the 60 days prior to the pricing date
13
Price Impact: Average Abnormal Returns of Underlying
Commodity Futures, Excluding S&P GSCI Roll Periods
Panel A: Individual commodities and commodity baskets
Proceeds ≥ $2 million
Days
Day 0
Day 1
[0,1]
Proceeds ≥ $5 million
Days
Day 0
Day 1
[0,1]
Proceeds ≥ $10 million
Days
Day 0
Day 1
[0,1]
0.22%
0.14%
0.37%
0.21%
0.19%
0.40%
0.31%
0.20%
0.51%
t-statistic
2.78
1.74
2.97
2.28
2.04
2.88
2.81
1.91
3.09
Number of returns > 0
234
239
254
169
179
193
128
130
149
Number of returns ≤ 0
224
219
204
172
162
148
124
122
103
0.3371
0.1873
0.0110
0.5857
0.1931
0.0085
0.4251
0.3297
0.0022
Average abnormal return
Probability under H0
Panel B: Individual commodities (excluding commodity baskets)
Proceeds ≥ $2 million
Days
Day 0
Day 1
[0,1]
Proceeds ≥ $5 million
Days
Day 0
Day 1
[0,1]
Proceeds ≥ $10 million
Days
Day 0
Day 1
[0,1]
0.23%
0.13%
0.37%
0.22%
0.18%
0.40%
0.33%
0.19%
0.52%
t-statistic
2.66
1.47
2.71
2.15
1.81
2.63
2.72
1.74
2.90
Number of returns > 0
208
213
224
152
160
172
117
117
136
Number of returns ≤ 0
203
198
187
157
149
137
114
114
95
0.4218
0.2449
0.0378
0.6335
0.2848
0.0265
0.4477
0.4477
0.0042
Average abnormal return
Probability under H0
14
Price Impact by Commodity Sector (Energy/Other)
Panel A: Energy Commodities
Average abnormal return
t-statistic
Number of returns >0
Number of returns ≤0
Probability under H0
Proceeds ≥ $2 million
Day 0
Day 1
Days [0,1]
0.31%
0.65%
0.96%
1.44
2.63
2.66
51
60
62
56
47
45
0.7190
0.1229
0.0608
Proceeds ≥ $5 million
Day 0
Day 1
Days [0,1]
1.14%
0.41%
0.72%
1.45
2.56
2.52
35
40
44
37
32
28
0.6380
0.2048
0.0382
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
0.75% 0.68%
1.43%
2.09
2.52
1.95
28
27
31
24
25
21
0.3389 0.4449
0.1058
Panel B: Other Commodities
Average abnormal return
t-statistic
Number of returns >0
Number of returns ≤0
Probability under H0
Proceeds ≥ $2 million
Day 1
Day 0
Days [0,1]
0.20%
-0.01%
0.18%
2.39
-0.14
1.60
183
179
192
168
172
159
0.2275
0.3744
0.0437
Proceeds ≥ $5 million
Day 1
Days [0,1]
Day 0
0.16%
0.05%
0.22%
1.74
0.58
1.67
134
139
149
135
130
120
0.5485
0.3129
0.0438
Proceeds ≥ $10 million
Day 1
Days [0,1]
Day 0
0.20% 0.08%
0.27%
1.89
0.78
1.84
100
103
118
100
97
82
0.5282 0.3619
0.0066
15
Do CLN Issues Convey Information?
• We reject the possible alternative explanation that
CLN issues convey information to the market because
futures contracts/ETF’s are a better trading vehicle
for an investor with information:
– Futures have high liquidity, low transactions costs,
embedded leverage, and no credit risk
– ETF’s have high liquidity, low transactions costs, and less
credit risk than CLN’s
– CLN’s have low liquidity (and are only available when an
issuer is selling a new issue), are costly to exit prior to
maturity, have high commissions and embedded fees, and
carry the credit risk of the issuer
• Implausible that CLN investors are sophisticated
enough to have valuable information about
commodity prices but so unsophisticated that they are
unaware of the advantages of using futures contracts
16
Do CLN Issues Convey Information?
• But, there are some fixed-income managers whose
mandates limit their investments to notes and bonds
• They are allowed to invest in CLN’s because CLN’s
are notes, but are not permitted to buy futures or
ETF’s
• Is it possible that some such fixed-income managers
have valuable information about future commodity
prices and trade on that information using CLN’s?
• If this is the case then the CLN issues might convey
information to the market.
17
Average Abnormal Returns for High-Commission
(≥1%) Issues, Excluding S&P GSCI Roll Periods
Panel A: Individual commodities and commodity baskets
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.29%
0.19%
0.48%
0.34%
0.20%
0.55%
0.45%
0.32%
0.76%
t-statistic
2.80
1.64
2.85
2.94
1.74
3.15
3.35
2.30
3.89
Number of returns > 0
120
113
124
86
80
91
64
58
71
Number of returns ≤ 0
Probability under H0
88
95
84
64
70
59
42
48
35
0.0157
0.1192
0.0034
0.0430
0.2313
0.0056
0.0204
0.1911
0.0003
Average abnormal return
Panel B: Individual commodities (excluding commodity baskets)
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.29%
0.15%
0.44%
0.37%
0.15%
0.52%
0.48%
0.25%
0.73%
t-statistic
2.48
1.16
2.31
2.84
1.18
2.72
3.28
1.69
3.42
Number of returns > 0
103
96
101
78
67
76
59
50
63
Number of returns ≤ 0
Probability under H0
76
83
78
54
65
56
37
46
33
0.0258
0.1849
0.0499
0.0224
0.4653
0.0489
0.0158
0.3798
0.0014
Average abnormal return
18
Average Abnormal Returns for Low-Commission (<1%)
Issues, Excluding S&P GSCI Roll Periods
Panel A: Individual commodities and commodity baskets
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.16%
0.09%
0.26%
0.11%
0.16%
0.27%
0.21%
0.10%
0.31%
t-statistic
1.37
0.81
1.45
0.80
1.19
1.30
1.27
0.67
1.26
Number of returns > 0
116
126
130
83
99
103
66
72
78
Number of returns ≤ 0
Probability under H0
134
124
120
108
92
88
80
74
68
0.8853
0.4748
0.2847
0.9702
0.3321
0.1555
0.8928
0.5980
0.2282
Average abnormal return
Panel B: Individual commodities (excluding commodity baskets)
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.18%
0.11%
0.29%
0.11%
0.18%
0.29%
0.21%
0.14%
0.35%
t-statistic
1.44
0.86
1.52
0.71
1.29
1.30
1.20
0.86
1.32
Number of returns > 0
105
118
123
76
93
96
60
67
73
Number of returns ≤ 0
Probability under H0
127
114
109
101
84
81
75
68
62
0.9346
0.4220
0.1967
0.9748
0.2739
0.1463
0.9159
0.5683
0.1948
Average abnormal return
19
Do CLN Issues Convey Information?
In order to believe that the price impact is due to
information conveyed by the fact of the CLN issue, you
have to believe that there are investors:
•ho are sophisticated enough to have information about
future commodity prices
•But are so unsophisticated that they chose to trade on
the information using illiquid, high-commission CLN’s
rather than low-commission CLN’s, liquid, lowcommission ETN’s, ETF’s, and commodity futures
20
Selection Bias?
• Issuer can cancel issue up until the close of trading on the
pricing date. Could the positive abnormal returns be due to the
fact that issues tend to be completed on days when commodity
prices go up?
• If this selection bias is important, it should also affect issues
based on diversified indexes. In particular, the price impact for
issues based on diversified indexes is an upper bound on the
magnitude of the selection bias. This upper bound is close to
zero.
• If the price impact is due to selection bias, it should be
independent of the issue size. But, we find that price impact is
increasing in the issue size.
21
Price Impact on Diversified Commodity Indexes
Proceeds ≥ $2 million
Day 0
Day 1 Days [0,1]
Proceeds ≥ $5 million
Day 0
Day 1 Days [0,1]
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
-0.04%
0.01%
-0.03%
-0.07%
0.04%
-0.02%
-0.03%
0.00%
-0.03%
t-statistic
-0.55
0.16
-0.28
-0.73
0.42
-0.19
-0.28
0.02
-0.18
Number of returns > 0
193
206
210
155
163
165
118
118
121
Number of returns ≤ 0
198
185
181
156
148
146
113
113
110
0.6192
0.1559
0.0783
0.5451
0.2137
0.1537
0.3962
0.3962
0.2553
Average abnormal return
Probability under H0
No evidence of any price impact around the pricing dates
of issues based on diversified commodity indexes
22
Small Issues
(Proceeds < $2 million and proceeds < $5 million)
Proceeds < $2 million
Proceeds < $5 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
-0.14%
-0.21%
-0.35%
0.07%
-0.09%
-0.02%
-0.60
-0.99
-1.02
0.51
-0.62
-0.09
Number of returns > 0
54
45
50
119
105
111
Number of returns ≤ 0
46
55
50
98
112
106
0.2421
0.8644
0.5398
0.0872
0.7064
0.3930
Average abnormal return
t-statistic
Probability under H0
No evidence of any price impact around the pricing dates
of small issues
23
Are the Price Impacts Temporary or Permanent?
Post-Issue Returns
Panel A: Individual commodities and baskets
Day relative to pricing date
Total Proceeds ≥ $2 million
Average abnormal return
t-statistic
Number of returns > 0
Number of returns ≤ 0
Probability under H0
Total Proceeds ≥ $5 million
Average abnormal return
t-statistic
Number of returns > 0
Number of returns ≤ 0
Probability under H0
Total Proceeds ≥ $10 million
Average abnormal return
t-statistic
Number of returns > 0
Number of returns ≤ 0
Probability under H0
0-1
2
3
4
5
0.37%
2.97
254
204
0.0110
0.17%
1.93
244
214
0.0877
0.13%
1.37
242
216
0.1214
0.12%
1.40
250
208
0.0276
0.05%
0.55
234
224
0.3371
0.40%
2.88
193
148
0.0085
0.17%
1.61
183
158
0.0968
0.10%
0.92
177
164
0.2579
0.11%
1.05
186
155
0.0521
0.06%
0.58
175
166
0.3325
0.51%
3.09
149
103
0.0022
0.10%
0.81
134
118
0.1724
-0.01%
-0.05
125
127
0.5749
0.05%
0.39
132
120
0.2442
0.07%
0.56
127
125
0.4749
24
Post-Issue Returns
Panel B: Individual commodities (excluding baskets)
Day relative to pricing date
Total Proceeds ≥ $2 million
Average abnormal return
t-statistic
Number of returns > 0
Number of returns ≤ 0
Probability under H0
Total Proceeds ≥ $5 million
Average abnormal return
t-statistic
Number of returns > 0
Number of returns ≤ 0
Probability under H0
Total Proceeds ≥ $10 million
Average abnormal return
t-statistic
Number of returns > 0
Number of returns ≤ 0
Probability under H0
0-1
2
3
4
5
0.37%
2.71
224
187
0.0378
0.20%
2.01
217
194
0.1389
0.13%
1.28
218
193
0.1182
0.14%
1.41
226
185
0.0242
0.06%
0.59
210
201
0.3466
0.40%
2.63
172
137
0.0265
0.21%
1.83
165
144
0.1276
0.10%
0.88
162
147
0.2129
0.13%
1.17
171
138
0.0343
0.06%
0.50
157
152
0.4100
0.52%
2.90
136
95
0.0042
0.14%
1.07
123
108
0.1785
-0.02%
-0.18
113
118
0.6534
0.07%
0.55
122
109
0.2149
0.06%
0.45
117
114
0.4477
25
Why Are There Price Impacts?
• Shleifer (1986) argues that there are downward
sloping demand curves due to differing opinions
about fundamental values.
• Singleton (2011) provides evidence about
disagreement and its importance in the oil market
• Petajisto (2009) provides a model in which the
pressure is borne by intermediaries, and applies his
model to explain the empirical evidence on index
inclusions and deletions.
• Price impacts are consistent with idea that demand
due to hedge trades shifts the demand curve to the
right
26
Include issues during “Goldman Roll”
Main results exclude issues in the “Goldman roll.” We find
similar results when we include issues that take place during the
Goldman roll.
Panel A: Individual Commodities and Baskets
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.08%
0.21%
0.29%
0.06%
0.21%
0.26%
0.16%
0.22%
0.39%
t-statistic
1.02
2.52
2.45
0.60
2.37
1.92
1.58
2.20
2.44
Number of returns > 0
265
281
290
189
210
218
141
149
166
Number of returns ≤ 0
273
257
248
214
193
185
153
145
128
0.6510
0.1607
0.0385
0.9024
0.2127
0.0554
0.7758
0.4306
0.0154
Average abnormal return
Probability under H0
Panel B: Individual Commodities (Excluding Baskets)
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.07%
0.20%
0.27%
0.05%
0.20%
0.25%
0.17%
0.22%
0.39%
t-statistic
0.83
2.17
2.07
0.50
2.11
1.67
1.51
1.98
2.24
Number of returns > 0
233
248
252
169
187
192
127
132
148
Number of returns ≤ 0
Probability under H0
247
232
228
194
176
171
138
133
117
0.7532
0.2468
0.1469
0.9139
0.2999
0.1469
0.7695
0.5489
0.0326
Average abnormal return
27
Mean-adjusted returns
Obtain similar results when we use the average return over 60day window leading up to the pricing date as the benchmark.
Panel A: Individual commodities and commodity baskets
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.19%
0.13%
0.33%
0.21%
0.18%
0.38%
0.26%
0.24%
0.50%
t-statistic
2.24
1.49
2.50
2.24
1.85
2.72
2.38
2.08
2.95
Number of returns > 0
252
244
257
188
184
195
142
135
150
Number of returns ≤ 0
Probability under H0
206
214
201
153
157
146
110
117
102
0.0177
0.0877
0.0050
0.0327
0.0795
0.0046
0.0253
0.1421
0.0015
Average abnormal return
Panel B: Individual commodities (excluding commodity baskets)
Proceeds ≥ $2 million
Proceeds ≥ $5 million
Proceeds ≥ $10 million
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
Day 0
Day 1
Days [0,1]
0.21%
0.11%
0.33%
0.22%
0.16%
0.37%
0.28%
0.23%
0.51%
t-statistic
2.26
1.16
2.28
2.16
1.50
2.42
2.38
1.89
2.81
Number of returns > 0
229
216
227
172
164
173
133
123
136
Number of returns ≤ 0
182
195
184
137
145
136
98
108
95
0.0116
0.1619
0.0191
0.0265
0.1529
0.0202
0.0125
0.1785
0.0042
Average abnormal return
Probability under H0
28
Relative Issue Size and Price Impact
Right-hand Side
Variable
Constant
RelativeIssueSize
Coefficient Estimates and t-statistics
(in parenthesis)
(1)
(2)
(3)
0.001
0.0013
-0.0003
(0.89)
(0.86)
(-0.08)
0.217
(3.03)
AbnormalVolume
0.0056
(1.78)
2-DayVolume
0.0057
(1.81)
2×AverageVolume
-0.0043
(-1.06)
Number of Observations
374
347
347
Dependent variable is the abnormal futures return over the twoday period Days[0,1]
29
Right-hand side variables:
Relative Issue Size = total proceeds divided by dollar open
interest of the two-nearest expiring futures contracts
Abnormal Volume = average volume in the reference
commodity’s front month futures contract during Days [0,1]
minus the trailing 60-day average daily volume, normalized by
the open interest in the two-nearest expiring futures contracts
2-DayVolume = volume during Days[0,1], normalized by the
open interest in the two-nearest expiring futures contracts.
AverageVolume = trailing 60-day average daily volume,
normalized by the open interest in the two-nearest expiring
futures contracts
30
Conclusion
• The hedge trades associated with CLNs have
statistically and economically significant price impacts
on the underlying commodity futures markets.
• The average two-day abnormal return around the
pricing dates of CLNs is 37 to 51 basis points.
• No evidence of reversals.
• These results support the view that non-information
based “financial” trading has an important impact on
commodity futures prices, i.e. it is evidence consistent
with the “financialization” of commodity futures
markets