HIGH PRICE VOLATILITY AND SPILLOVER EFFECTS IN ENERGY MARKETS Aaron Sing Gradate Stdent Department of Agricltral and Applied Economics Te Uniersity of Georgia Atens, GA 3060-7509 Email: singaa@ga.ed Berna Karali Assistant Professor Department of Agricltral and Applied Economics Te Uniersity of Georgia Atens, GA 3060-7509 Telepone: (706) 54-0750 Fax: (706) 54-0739 Email: bkarali@ga.ed Octaio A. Ramirez Professor Department of Agricltral and Applied Economics Te Uniersity of Georgia Atens, GA 3060-7509 Telepone: (706) 54-481 Fax: (706) 54-0739 Email: oramirez@ga.ed Selected Paper prepared for presentation at te Agricltral & Applied Economics Association s 011 AAEA & NAREA Joint Annal Meeting, Pittsbrg, Pennsylania, Jly 4-6, 011. Copyrigt 011 by Aaron Sing, Berna Karali, and Octaio A. Ramirez. All rigts resered. Readers may make erbatim copies of tis docment for non-commercial prposes by any means, proided tis copyrigt notice appears on all sc copies.
Hig Price Volatility and Spilloer Effects in Energy Markets Abstract We analyze te time-arying olatility in crde oil, eating oil, and natral gas ftres markets by incorporating canges in important macroeconomic ariables and major political and weater-related eents into te conditional ariance eqations. We allow asymmetric responses to random distrbances in eac market as well as to good and bad economic news in te oerall economy. We also inestigate weter tere are spilloer effects among tese energy markets. A bi-directional olatility spilloer effect is fond between eating oil and natral gas markets. Among te macro ariables considered te spread between te 10-year and -year Treasry constant matrity rate is fond to ae a positie relationsip between te olatilities of all commodities. Te eents tat ad a major impact on te olatilities of energy commodities inclde te September 11 t terrorist attacks, rricane Katrina, and te 008 U.S. financial crisis. Te teory of storage is not spported in any of te tree commodities. Seasonality and day-of-te week effects are fond for all tree commodities. Key words: Asymmetric socks, energy markets, GARCH, oil, spilloer effects, olatility
Hig Price Volatility and Spilloer Effects in Energy Markets Introdction Since te smmer of 008, energy ftres prices ae experienced ig olatility inclding a dramatic drop in oil prices from a record ig leel to less tan alf te ale in jst a few monts. In conjnction, inestors ae been faced wit ig leels of ncertainty abot eqity markets and te direction of economic policy, reslting in iger olatility in commodity prices. As a reslt of tis olatility, te Cicago Mercantile Excange Grop introdced new crde oil olatility index ftres and options contracts based on olatility indexes calclated by te Cicago Board of Options Excange to elp prodcers and traders to track and trade on te olatility in crde oil. Te index calclations se prices from te excange's actie and liqid options on ftres markets to create new and effectie measres of expected olatility. Tis introdction of new contracts presents a nmber of new opportnities for edging/managing/speclating price risk, bt also presents new callenges becase of te difficlty of measring expected olatility. To measre expected olatility, it is ery important to nderstand te relationsip between different energy prodcts, teir price determinants, and te nderlying factors beind teir price flctations. Crde oil is a large component of prodction costs for eating oil and gasoline, and terefore flctations in te price of crde oil sold reslt in corresponding flctations in eating oil and gasoline prices. In fact, olatilities of crde oil, eating oil, and gasoline are fond to be igly correlated (Pindyck, 001). On te oter and, crde oil is a close sbstitte for natral gas as an energy sorce, and ts
crde oil price flctations sold also affect natral gas prices (M, 007). Frter, olatility transmission between oil and natral gas markets as been fond (Ewing et al., 00). Conseqently, it is important to analyze all tese markets simltaneosly to determine te factors beind teir price olatility. Price determinants inclde demand and spply factors. Weater plays an important role in te demand side of energy markets. Colder tan normal temperatres in winter and otter tan normal temperatres in smmer can increase natral gas demand and ps p prices. Demand for crde oil and eating oil peaks in winter as well. Economic growt reslts in increased demand for goods and serices from te commercial and indstrial sectors, and terefore generates an increase in demand for bot crde oil and natral gas. On te spply side, OPEC decisions abot prodction and prices, political eents, storage leels, and natral eents are among te determinants of energy prices. Macroeconomic factors affecting energy prices ae been also stdied in te literatre. Some of te ariables fond to affect energy prices inclde bilateral excange rates, price indices, monetary aggregates (Zagaglia, 010); conenience yield (Lin and Dan, 007); Treasry bill yields, eqity diidend yields, and jnk bond premims (Bessembinder and Can, 199). Te goal of tis stdy is to simltaneosly estimate te price olatility in energy markets wile acconting for spilloer effects across different commodities. Daily settlement prices of te nearby crde oil, eating oil, and natral gas ftres contracts are inclded in te empirical analysis. Frter, macroeconomic indicators, inclding percentage canges in consmer price index, indstrial prodction index, and inentory leels as well as te spread between te ten- and two-year constant matrity Treasry 3
bonds are sed to test weter tese ariables affect energy price olatility. Tese macroeconomic factors elp to examine olatility trends dring different periods of te bsiness cycle. For instance, olatility in oil ftres recently fell sligtly as indstrial prodction rose more tan expected in December 010. Ts, or reslts can assist market participants in better nderstanding wic direction olatility in energy markets go wen te leels of tese macroeconomic factors cange. We also analyze te impact of major political eents, sc as canges in OPEC policies, on energy price olatility. Tis is important becase OPEC recently released a statement in late winter of 010 saying tat maintaining crrent oil prices in a range of $80-$100 wold be ideal. In April 011, te ICE Brent Crde oil srpassed tis leel by far, closing at arond $15 for a few consectie days. Major natral eents like rricane Katrina is sed to accont for spply socks. To captre te impact of weater montly dmmy ariables are sed. We se a mltiariate GARCH model to simltaneosly estimate, te spilloer effects across energy markets as well. Or stdy answers te following researc qestions: Does olatility in crde oil prices ae a spilloer effect on te olatilities of natral gas and eating oil? Wic economic and natral factors most explain olatility in energy markets? Is te teory of storage spported in energy ftres markets? Are tere leerage effects, i.e. asymmetric response to positie and negatie socks? Literatre Reiew Oer te last few decades, te United States as seen sbstantial increases in energy se and dependence on oter contries, casing energy prices to rise. Crde oil and gasoline 4
prices are particlarly of great concern. Oil as become te most traded commodity worldwide as bot deeloped and deeloping economies ae seen growt and increased demand for energy. Crde oil as an effect on oseolds as its price wold affect gasoline and fel prices and ts wold alter consmers decisions on trael and prcases of related items, sc as atomobiles. For example, Kilian (008) sowed tat an nanticipated energy price increase of abot one percent cased a decrease of almost te same magnitde in te prcases of motor eicles and parts. Frtermore, te real consmption of domestic atomobiles as decreased compared to foreign atomobiles. Tis was becase te U.S. consmers typically perceie domestic atomobiles 1 as less fel-efficient. Energy price socks affect not only consmer side, bt also non-residential inestment or bsiness consmption. Oil can be seen as an intermediate good tat is sed in te prodction of final goods. Ts, if oil prices are ig ten firms will lower teir prodction and tis will, in trn, case a contraction in te economy. Bot crde oil and natral gas prices ae been more olatile tan almost all prodcer prodcts from 1945 to 005 (Regnier, 007). Oil price ncertainty as been fond to ae negatie and significant effect on te aerage growt rate of real economic actiity (Raman and Serletis, 010) and ence sold be addressed careflly by policy makers. Te effect of oil price ncertainty on non-residential inestments as been docmented by Elder and Serletis (010). Domestic mining expenditres were fond to decrease sbstantially wen tere was a decrease in oil prices. Howeer, mining expenditres were fond to increase ery little wen tere was an increase in oil prices. One ting to note is tat in tis stdy, 1 Te U.S. Brea of Economic Analysis defines domestic cars as cars assembled in te U.S., Canada, and Mexico. 5
oil price ncertainty was ery low dring te period of 00-008 een tog oil prices were on a continal rise dring tat time. Tis can elp to explain wy tere weren t more recessionary times een tog te continos rise in oil price wold sggest tat economic downtrns wold be more prealent. In earlier stdies, oweer, Hamilton (1983, 004) sowed tat all post-war U.S. recessions were preceded by increases in oil prices. Additionally, in is 1983 paper, e fond tat recessions typically lag large increases in crde oil prices by tree to for monts. Te only exception was te recession of 1960-1961. Te United States officially entered into a recession in December 007 and exited in Jne 009. Since ten, te U.S. economy as been experiencing wat some economists term as slggis growt and tis rate as been reised to 1.9 percent in te first qarter of 011. Dring tis time period crde oil prices it a little oer $140 per barrel, wic was te all time ig. Since ten, crde oil price as sbstantially dropped to less tan alf of its all time ig, and as of early Jly 011 trades arond $96 per barrel for te Agst contract. In te smmer of 008 tere was a sbstantial psrge in commodity prices, wic emerged after te Federal Resere started being open abot additional expansionary monetary policy (Lanman and Miller, 008). Sc a price increase is important to note, since te U.S. was in te midst of an economic downtrn and intitiely one wold expect mc lower oil and energy prices de to decreased demand. One example wold be te Asian financial crisis in 1997, wic started by te dealation of te Bat and spread trog oter Asian contries. Tis cased a major slowdown in energy demand by deeloping contries and as a reslt oil price dropped to 6
$8 per barrel by te end of 1998. Olowe (010) sowed tat te Asian financial crisis did ae an impact on crde oil price retrns wereas te global crisis of 008 did not. Strctral canges, sc as OPEC s pricing cange in April 1999, ae been sown to ae an effect on oil price olatility (Lee and Zyren, 007). OPEC redced qotas to boost oil prices after teir low leels seen in 1997. As recent as Janary 011, OPEC as come nder scrtiny as many qestioned if tey wold alter teir prodction in te wake of oil prices reacing te pper bond of te $80-$100 price range tat OPEC deemed satisfactory. Crde oil and natral gas are sbstittes as inpts in prodction, or as sorces of energy. Tis relationsip is key to bot consmers and prodcers of energy prodcts as te price dynamics of bot wold dictate weter to increase or decrease inentories, or een alter te rates of sbstittion between te two. As sc, tere ae been stdies tat analyzed cointegration of bot natral gas and crde oil prices, and natral gas and eating oil prices. Serletis and Herbert (1999), for instance, stdied daily Henry Hb and Transco Zone 6 natral gas prices, fel oil, and power prices. Tey fond tat tese tree fel prices (except for power price) were all cointegrated. Serletis and Rangel-Riz (004) fond a decopling in daily natral gas and crde oil prices from Janary 1991 to April 001. Tis meant tat tere were no common or codependent cycles between te prices of natral gas and crde oil. In a later stdy, sing an error-correction model Brown and Ycel (008) fond natral gas and crde oil prices to be cointegrated in te long rn. Howeer, tey also fond tat in te sort-rn, natral gas prices cold deiate from crde oil prices becase of factors sc as inentory leels and weater. Bilding on te literatre on te relationsip between natral gas and crde oil, Hartley et al. 7
(008) fond tat crde oil and natral gas ae an indirect relationsip ia eating oil. Additionally, in agreement wit Brown and Ycel (008) tey fond tat factors sc as weater, rricanes, and inentory leels all ad significant effects on te sort rn relationsip between natral gas and crde oil prices. Tis is especially important for commercial sers of bot commodities, as tey may want to cange teir inentories accordingly. Commodity markets exibit seasonality as sown by Senaga et al. (008). Volatility of natral gas was fond to be greater in winter tan it is in smmer. Intitiely, one wold expect iger olatility of natral gas prices dring winter as bsinesses and oseolds increase teir demand for eating and energy se. Weater and storage of natral gas also ties into seasonality, as tere are different weater patterns dring different seasons and it is necessary to store a commodity wen tere is scarcity. Natral gas and crde oil spot prices ae been sown to ae a negatie correlation wit inentories (Geman and Oana, 009). Tis correlation increases sbstantially dring winter monts. Tis eidence is ital to bsinesses tat experience increases in eat and oter energy demand in winter. Volatility spilloers and asymmetries are important to inestors in order to bild an optimal portfolio. Spilloer effects are indirect externalities tat arise from and are cased by some oter penomenon i.e., a cange or sock in one sector tat as an effect tat carries oer to anoter sector. Cang et al. (010) fond spilloer effects from Brent crde ftres retrns to Brent crde spot and forward retrns. Additionally, tere were spilloer effects from WTI ftres retrns to Brent spot retrns and from Brent spot to WTI spot retrns. Tese were all in one direction as stated and tere were not many 8
spilloer effects tat moed in bot directions. Frtermore, spilloer effects were fond in crde oil, gasoline, and eating oil markets (Hammode et al., 003) particlarly in nearby ftres contracts and spot prices. Volatility transmission was fond in spot, onemont, and tree-mont prices for WTI crde oil and was more prealent tan mean retrns transmission. Bot gasoline and eating oil ad olatility transmission from te spot prices to te one- and tree-mont prices as well, were tey were eac different for mean retrns transmission. Ewing et al. (00) docmented olatility transmission between crde oil and natral gas markets, sowing tat oil olatility depended on past olatility of natral gas, wereas natral gas olatility depended on nexpected eents. Or stdy bilds on tis extensie literatre and focses on te economic determinants of olatility in crde oil, eating oil, and natral gas ftres markets as well as on te spilloer effects across tese markets. Model Te atoregressie conditional eteroskedasticity (ARCH) model was first introdced by Engle (198) and as been widely sed to measre olatility in financial markets. In tis model te ariance of te crrent error term is a fnction of te sqared past error terms. Tis model was later generalized by Bollersle (1986) to inclde lagged ales of te ariance as well and called te generalized atoregressie conditional eteroskedasticity (GARCH) model. GARCH models are fond to be sefl in explaining stock price distribtions (Bollersle, 1987; Bollersle et al. 1988; Frenc et al., 1987; Balliie and DeGennaro, 1990). It as been sown tat commodity ftres prices also exibit timearying olatility and can be effectiely stdied sing GARCH models (Baillie and 9
Myers, 1991; Myers, 1991; Myers and Hanson, 1993; Yang and Brorsen, 1993; Goodwin and Scnepf, 000). We adopt te mltiariate GARCH-BEKK model deeloped by Engle and Kroner (1995) in or stdy and modify it to inclde exogenos ariables tat migt ae an impact on te conditional olatility. We measre te daily retrn from olding a ftres contract on day t as r t 100 ln Ft ln Ft 1, (1) were F t is daily settlement price of te ftres contract on day t. Te mean eqation of daily retrns is ten defined as a fnction of its past ales and a random distrbance term. Denoting te ector of mean retrns by R t, te mltiariate GARCH in matrix form is gien by: R t p R, MVN (0, H ), () i1 ti t t ~ t were R t is a 3x1 ector consisting of r t s of eac commodity, p is te order of atoregressie process, and t is te distrbance ector. Te conditional coariance matrix of te distrbance term is ten gien by: H t CC A t1 t 1A B H t1b D t 1t 1D G GX t, (3) were t1 = t 1 I <0, wic replicates te ector t1 wit positie elements zeroed ot. H t is a 3x3 symmetric matrix wit ariances on te diagonal and coariances off te diagonal. C is a 3x3 lower trianglar matrix of constants, A is a 3x3 matrix of ARCH parameters, B is a 3x3 matrix of GARCH parameters, D is a 3x3 matrix tat measres 10
asymmetric ARCH effects and G is 3x3 lower trianglar coefficient matrix on te exogenos ariables X t. Te matrices are as follows:,,, 0 0 H t =,,,, C = 0, A =,,,, 0 0 B =, D =, G = 0, were te sbscripts 1,, and 3 represent, respectiely, crde oil, eating oil, and natral gas. Matrix maniplation yields te conditional ariance eqations sown as:, c 11 b 11 11, t1 d c d 1 c b 31 1, t1 a 11 1 1, t1, t1 11 1, t1 b d a 31 33, t1 d 1, t1 b b a 11 31 1, t1 3, t1 31 3, t1 11 1 1, t1 d d a b b a 11 31 13, t1 1 31, t1 3, t1 11 1 1, t1 ( g, t1 b 11 a 1 31 3, t1 g b 1 a g 31 d ) X t 11 31 1, t1 3, t1 11 1, t1 a d a 1, t1 1 31, t1 d 3, t1 31 3, t1 (4), c b c 1 11, t1 d 3 d a b 1 1, t1, t1 1 1, t1, t1 a b d 3, t1 33, t1 d a 3 b 3, t1 b 1 3 1, t1 3, t1 a 1 1, t1 d b d a b 1 1, t1, t1 1 3 13, t1 3, t1 3, t1 a b g a b g 3 X t 1 3 1, t1 3, t1 3 3, t1 d a 1 1, t1 a d 3, t1 3, t1 (5),, t1 d 3 3, t1 c 33 b a 13 11, t1 d d 13 1, t1 b 3 a, t1 3 13 3 1, t1, t1, t1 b d a 33 33, t1 d 33 3, t1 b a b 13 33 1, t1 3, t1 a d 13 3 1, t1 b d 13 3 1, t1, t1 b a 13 33 13, t1 3 33, t1 3, t1 a b g 33 X t b 13 33 1, t1 3, t1 a 3 33 3, t1 d a 13 1, t1 3 33, t1 3, t1 d 3, t1 d 33 3, t1 (6) 11
Data We stdy tree selected energy ftres contracts tat are traded on te New York Mercantile Excange (NYMEX): crde oil, eating oil, and natral gas. Ligt sweet crde oil (WTI) ftres contracts ae expiry dates in eery mont of te year and are traded ntil te tird bsiness day prior to te 5 t calendar day of te mont preceding te deliery mont. Standard contract size is 1,000 barrels and price is qoted as U.S. dollars and cents per barrel. Heating oil ftres contracts also ae expiry dates in all monts of te year and are traded ntil te last bsiness day of te mont preceding te deliery mont. Heating oil contract size is 4,000 gallons and price is qoted in U.S. dollars and cents per gallon. Natral gas (Henry Hb) ftres contracts too ae expiry dates in all monts of te year and terminate trading tree bsiness days prior to te first day of te deliery mont. Eac contract stands for 10,000 million Britis termal nits (mmbt) and qoted in U.S. dollars and cents per mmbt. We constrct price series for all tree commodities by rolling oer teir first nearby contracts on te 15 t day of expiration mont (te mont preceding te contract mont). Ftres price data are obtained from Commodity Researc Brea and Datastream proided by Tomson Reters. Or sample coers te period from Febrary 1, 1994 to Febrary 4, 011. We stdy te impact of macroeconomic ariables as well as major political or natral eents on te olatility in energy markets. To tis end, we se percentage canges in Consmer Price Index for All Urban Consmers: All Items and Indstrial Prodction Index, and te spread between te 10-year and -year Treasry Constant Matrity Rate obtained from Arcial Federal Resere Economic Data (ALFRED). ALFRED is released by te Economic Researc Diision of te Federal Resere Bank of 1
St. Lois and contains data on major economic ariables aailable at te time of te release (witot any reisions made). All tese ariables are recorded montly. We interpolate tese montly series ia a step fnction to obtain daily series in order to se wit or daily ftres retrns. Becase inentories play an important role in stabilizing demand and spply socks for storable commodities we also inclde inentory data in or olatility analysis. Inentory data for all commodities are obtained from te U.S. Energy Information Administration (EIA). For crde oil, we se te Weekly U.S. Ending Stocks of Crde Oil series stated in tosand barrels. For eating oil, we se Weekly U.S. Ending Stocks of Distillate Fel Oil stated in tosand barrels; and finally for natral gas, we se te series Weekly Lower 48 States Natral Gas Working Undergrond Storage stated in billion cbic feet. We compte percentage canges in inentories from one week to te next for eac commodity and interpolate te reslting weekly series ia a step fnction to obtain daily series to matc te freqency wit or price data. Finally, dmmy ariables are sed to accont for te days of te week, calendar monts, and political and weater-related eents tat affect te world price of crde oil. Te day of Friday and te mont of December are sed as base categories and ts teir effects are sown in te intercept. Te discssion of te inclded eents follows. (1) Te Asian economic crisis tat lasted from Jly 1997 to Febrary 1998. Tis began wit te collapse of te Tailand Bat and spread to many Asian contries. By late 1998, crde oil was priced at $8 per barrel and OPEC saw a need to ae a sift in policy to restore oil prices to iger leels. For tis eent, or ariable ASNFC takes te ale of one on te dates between Jly 1, 1997 and Febrary 8, 1998, and zero oterwise. () Te pledge by 13
OPEC and non-opec contries to ct otpt by a combined of.104 million barrels per day. Tis eent was a spply sock and ence directly increased te crde oil world price. Te ariable OPEC takes te ale of one on te dates between Marc 3, 1999 and Marc, 000, and zero oterwise. (3) Te terrorist attacks on September 11, 001. As a reslt of tese attacks all market operations were alted and ten resmed on September 17, 001. Howeer, tis eent canged te relationsips wit te Middle East permanently. Te ariable SEP11 takes te ale of one on te date September 11, 001 and tereafter. (4) Te U.S. inasion of Iraq on Marc 19, 003. It was reported tat Iraq ad lanced missile attacks on Kwait bt tere was no effect on any oil prodction facilities reported (Te Financial Express, 003). Or ariable for tis eent named IRQINV takes te ale of one on te dates between Marc 19, 003 and April 17, 003. (5) Hrricane Katrina it te U.S. Glf coast on Agst 9, 005. Katrina was te costliest rricane eer to it te U.S. Glf coast and te sixt strongest Atlantic rricane eent. Not only did tis affect crde oil prices, bt also natral gas prices. Katrina damaged or destroyed 30 oil platforms. Additionally, abot nine refineries were forced to close down for te following six monts, and te total loss in oil prodction in te Glf coast was acconted for 4% of annal prodction. To accont for tis major eent, te ariable KTRN takes te ale of one on te dates between Agst 9, 005 and Febrary 8, 006, and zero oterwise. (6) Te U.S. financial crisis became prealent on September 15, 008 wen te major inestment bank Leman Broters annonced tat it will be filing for bankrptcy. Tis cased many ripple effects as credit dried p in te financial markets, casing a credit constraint for firms and consmers. Tis wold ten case a sbstantial decrease in demand for crde oil, gasoline, and oter energy 14
commodities. For tis eent, or ariable LEHMN takes te ale of one on te dates between September 15, 008 and Jne 30, 009. Table 1 presents descriptie statistics of te daily ftres retrns and macro ariables employed in te empirical analysis. Table sows te nit root test reslts for ftres price series. As can be seen in te table, bot te leels and te logs of ftres prices in all markets contain a nit root, tat is, tese series are nonstationary. Howeer, we can reject te existence of a nit root for te retrn series, compted as te differences of log ftres prices. Empirical Reslts For all tree commodities we estimate a mltiariate GARCH BEKK model wit lagged retrns inclded in te mean eqations. Conditional ariance eqations inclde ARCH, asymmetric ARCH, and GARCH parameters as well as exogenos ariables discssed earlier tat migt ae an impact on olatility. In order to test weter pward canges in economic ariables affect olatility differently tan downward canges do, we inclde indicator ariables for negatie canges in consmer price index (CPI), indstrial prodction index (IP), and inentories (INV). Table 3 presents te coefficient estimates and teir p-ales for te ariance eqations gien in (4)-(6). Crde Oil Te mean eqation reslts sow a constant retrn of 0.07 in crde oil ftres. Te first tree lagged retrns are significant, wit a positie coefficient on te first lag and negatie on te oters. Te constant conditional ariance is 10.4. Te ARCH parameter 15
of 0.0 implies tat positie distrbances (socks, news) to crde oil increase conditional ariance by tat amont. Howeer, past positie socks to eating oil and natral gas olatility were fond to be insignificant. Te asymmetric ARCH coefficient for crde oil is 0.14, wic means tat past negatie distrbances to crde oil increases te crrent conditional ariance by 0.16. Additionally, negatie socks in eating oil markets are fond to increase te conditional ariance of crde oil by 0.13, sowing spilloer effects from eating oil to crde oil market. Te GARCH parameter for crde oil is 0.91, sowing tat crde oil olatility in te past period as a large effect on olatility in te crrent period and is igly persistent. Lagged ariance of natral gas retrns is also fond to increase te crrent ariance of crde oil retrns bt by a ery small amont. Conditional ariance reslts sow tat te strctral cange by OPEC, te September 11 t terrorist attacks, te U.S. inasion of Iraq, rricane Katrina, and te 008 financial crisis wic was eleated by te annoncement of Leman Broters to file bankrptcy reslted in an increase in crde oil price olatility. Tese eents increased te conditional ariance by 1.18, 5.03, 1.73, 4.88, and 5.1 percent respectiely. For te macro ariables, positie and negatie percent canges in CPI and eating oil inentories, te spread between te 10-year and -year Treasry bonds, and negatie percent canges in natral gas inentories all ae significant effects on te conditional ariance of crde oil retrns. For a one-percent increase in CPI, te conditional ariance increases by 1.04 percent wile for a one-percent decrease in CPI, te ariance decreases by 9.08 percent. Tis cold be te reslt of bsinesses seeing an opportnity to expand or prodce more on lower energy prices, ts driing p te demand. Increased demand can be one factor tat leads to iger price olatility, altog it is not te only factor. As te spread 16
between te 10-year and -year interest rates increases by one percent, crde oil ftres retrn ariance increases by 0.13 percent. Interestingly, te canges in crde oil inentories are not statistically significant. Howeer, eating oil inentories are fond to affect crde oil ariance. For a one-percent increase in eating oil inentories, te conditional ariance of crde oil increases by 0.04 percent and for a one-percent decline it decreases by 0.07 percent. Similarly, for a one-percent decrease in natral gas inentories, te conditional ariance of crde oil decreases by 0.0 percent. Tese inentory effects are pzzling becase one wold expect iger crde oil olatility wit lower eating oil or natral gas inentories. Crde oil ariance is fond to be iger on Mondays and Trsdays compared to te base category of Fridays. Interestingly olatility on Wednesdays is not iger tan Fridays een tog te U.S. Energy Information Administration releases te weekly inentory report on Wednesdays. Higer olatility on Mondays can be te reslt of any major news tat may ae taken place dring te weekend. All montly dmmy ariables except for Janary are fond to be significant, sowing iger olatility compared to December. Heating Oil Heating oil ftres ae a constant retrn of 0.08. Atocorrelation in te retrns is fond only in te first tree lags. Wile te coefficient on te first lagged retrn is positie, te coefficients on te second and tird lagged retrns are negatie. Te constant conditional ariance is.91. Te ARCH parameter is 0.07 and statistically significant. Tere is significant bt ery small spilloer effects from te crde oil market. Te asymmetric ARCH term for eating oil is 0.03, wic sggests tat past negatie news in te eating 17
oil market increases te crrent conditional ariance by 0.11. Bot positie and negatie distrbances to crde oil markets are also fond to increase te ariance of eating oil retrns by 0.01 and 0.04, respectiely. Te GARCH parameter is 0.8, sowing a ig leel of persistence. Additionally olatility spilloer effect from bot crde oil and natral gas markets to te eating oil market is fond bt te magnitde is small. Te Asian financial crisis and te strctral cange by OPEC are fond to ae significant impact on te conditional ariance of eating oil ftres. Te eating oil ariance increased by.66 de to Asian financial crisis in 1997, and increased by 0.56 after OPEC s qota cts in 1999. Among te macro ariables, bot positie and negatie canges in indstrial prodction index and eating oil inentories, positie canges in CPI and te spread between te 10-year and -year Treasry bonds are fond to be statistically significant. A one-percent increase in CPI increases te conditional ariance by 4.11 percent. A one-percent increase in indstrial prodction increases te conditional ariance by 0.66, wereas a one-percent decrease in indstrial prodction lowers it by 3.9. A one-percent increase in eating oil inentories raises te conditional ariance by 0.04 and a-one percent decrease cases te conditional ariance to decrease by 0.16. Tis is in contrast to wat te teory of storage predicts. As te spread between te 10-year and -year Treasry bonds increases by one percent te conditional ariance increases by 0.06. In terms of seasonality, only te monts of September and Noember exibit statistically iger olatility tan te mont of December. Tere are also significant dayof-te-week effects on olatility on Mondays and Wednesdays, wit bot being iger tan te olatility on Fridays. Becase eating oil inentory leel reports are released by te U.S. Energy Information Administration along wit te crde oil inentories, it is 18
expected to ae iger olatility on Wednesdays. Howeer, te same Wednesday effect was not fond for crde oil. Natral Gas Te constant retrn for natral gas ftres is 0.06 bt it is insignificant. Atocorrelation is fond in te first, te second, and te fift lags of retrns. Te ARCH parameter of 0.05 sggests tat past socks in natral gas markets do increase te crrent ariance by tis amont. Interestingly, tere were no significant asymmetric ARCH effects. Te GARCH parameter is 0.13. Unlike crde oil and eating oil, tis is significantly smaller and sggests tat olatility in te natral gas market is not as persistent as in te oter two markets. Frter, tere is a olatility spilloer effect from eating oil to te natral gas market (0.06). Tis is te largest olatility spilloer effect fond in any of tese energy markets and cold be de to te fact tat eating oil and natral gas are sbstittes for residential and commercial eating. Among te eents considered only rricane Katrina and te 008 financial crisis significantly increased te conditional ariance of natral gas by 8.94 and 8.9, respectiely. As for te macro ariables, a one-percent increase in te spread between te 10-year and -year Treasry bonds cases te conditional ariance of natral gas to increase by 0.06. Only pward canges in crde oil inentories and negatie canges in eating oil inentories are significant. Accordingly, te conditional ariance of natral gas increases by 0.49 for a one-percent increase in crde oil inentories and decreases by 0.05 for a one-percent decrease in eating oil inentories. Interestingly, te canges in natral gas inentory leels do not affect te olatility of natral gas ftres retrns. All 19
weekdays are fond to exibit iger olatility tan Fridays. Volatility in te monts of Marc trog Jne is fond to be iger tan in December. Te bottom part of table 3 sows model diagnostic tests. Te loglikeliood fnction ale is -5,619.9. Te Ljng-Box Q statistics sow tat we cannot reject te independence of te tree standardized residal series obtained from tis mltiariate GARCH model at 10%. Tis sows tat tere is no atocorrelation left in te residals and te model fits te data well. We also performed a likeliood ratio test to see weter te exogenos ariables inclded in te ariance eqations add any ale to te model. It is seen from te likeliood ratio test statistic and its p-ale tat we can reject te model wit no exogenos ariables. Conclsions Tis stdy inestigates te determinants of ig price olatility in energy ftres markets, namely crde oil, eating oil, and natral gas, wile acconting for asymmetric effects of news and possible spilloer effects across te markets. Frter, it analyzes te impact of major political and weater-related eents and te main macroeconomic ariables on te olatility in tese markets. Varios spilloer effects were fond in eac market, wit some being bidirectional and some being nidirectional. Heating oil is fond to be affected by te random socks in its own market and in te crde oil market. Heating oil is a by-prodct of crde oil and terefore one wold expect any sock in te crde oil market to ae an effect on eating oil olatility. Volatility transmission from natral gas to te crde oil market is fond. Tere is eidence of bi-directional olatility spilloer effects between 0
eating oil and natral gas, wic is expected as tey are sbstittes For asymmetric effects, tere was eidence of bi-directional spilloers between crde oil and eating oil. Te impact of negatie socks in te eating oil market on crde oil ariance is for times larger tan te impact of negatie socks in te crde oil market on eating oil ariance. Volatility in energy markets is fond to cange in response to major eents. Te Asian financial crisis only increased te olatility of eating oil retrns. OPEC s qota cts in 1999 increased te olatility of bot crde oil and eating oil, as one wold expect, since tis was a direct spply sock. Te terrorist attacks on September 11, 001 and te Iraq inasion in 003 increased only te olatility of crde oil. Een tog Iraq is part of te OPEC and its crde oil prodction is not conted, fears of destrction to oil facilities in te Middle East ae probably cased te increased olatility. Hrricane Katrina dramatically increased bot crde oil and natral gas olatility. Te reason for sc a ig increase in olatility is tat most of te crde oil and natral gas prodction facilities in te U.S. are sitated in te Glf Coast region. Among te macroeconomic ariables considered, te spread between te 10-year and -year Treasry bonds and negatie canges in eating oil inentories affect te olatility of all tree commodities. An increase in te Treasry bond spread implies a steeper yield cre, were te economy is expected to improe qickly. Since tese commodities are inpts for bsinesses and teir respectie prices are correlated wit te economy s performance, ten faster economic growt wold lead to greater demand and possibly iger price olatility. Positie canges in CPI ae te strongest effect on eating oil olatility and tis can be becase tis commodity is te most directly 1
consmed energy prodct by consmers. One interesting reslt was tat neiter crde oil nor natral gas olatility is affected by teir own inentory canges. A decrease in eating oil inentories is expected to increase crde oil olatility as eating oil is deried from crde oil and a sortage in eating oil wold increase te demand for crde oil. Similarly, an increase in demand for natral gas wold arise as a reslt of eating oil sortage. Howeer, bot crde oil and natral gas conditional ariance decreased wit a decrease in eating oil inentories. Crde oil market is fond to exibit strong seasonality wit iger olatility from Febrary trog Noember compared to December. Seasonality in oter markets is not as strong as in crde oil. In terms of daily patterns, reslts ary among commodities. Wile natral gas olatility is iger on all oter weekdays compared to Fridays, crde oil olatility is iger on Mondays and Trsdays and eating oil olatility is iger on Mondays and Wednesdays. Tis is interesting as te U.S. Energy Information Administration inentory reports are sally released on Wednesdays, so one wold expect all commodities to ae similar weekday patterns. Weekly natral gas inentory reports by te U.S. Energy Information Administration are issed on Trsdays, so iger olatility on tat day is expected. In recent monts, tere ad been mc debate on weter te U.S. Federal Resere sold contine its $600 billion qantitatie easing program, termed QE, wic ended on Jne 30 011. One impact of contining sc a program wold be a possible increase in inflation trog increasing te monetary base. As or reslts sow canges in consmer price index ae te strongest effect on crde oil and eating oil olatility. Ts, tis can ery well case major discomfort for consmers. Additionally, if policy
makers want to crb olatility in energy prices ten lowering te spread between longand sort-term interest rates interest rates can be of se as te Treasry bonds spread is fond to be positiely related to te olatility of all tree commodities. Lowering te spread can be acieed eiter increasing te sort-term interest rates or lowering te long-term rates. Tis wold not be ery easy task as many financial firms crrently depend on lower interest rates for many day-to-day operations in financial and commodity markets. In addition to adeqate monetary policy, reglations are ery mc necessary to be created and/or enforced in order to preent anoter financial calamity, as crde oil and natral gas olatilities were igly affected by te 008 U.S. financial crisis. 3
References Baillie, R.T., and R.P. DeGennaro. 1990. Stock Retrns and Volatility. Jornal of Financial and Qantitatie Analysis 5():03-14. Baillie, R.T., and R.J. Myers. 1991. Biariate GARCH Estimation of te Optimal Commodity Ftres Hedge. Jornal of Applied Econometrics 6:109-14. Bessembinder, H, and K. Can. 199. Time-Varying Risk Premia and Forecastable Retrns in Ftres Markets. Jornal of Financial Economics 3():169-193. Bollersle, T. 1986. Generalized Atoregressie Conditional Heteroskedasticity. Jornal of Econometrics 31(3):307-37. Bollersle, T. 1987. A Conditionally Heteroskedastic Time Series Model for Speclatie Prices and Rates of Retrn. Te Reiew of Economics and Statistics 69:54-547. Bollersle, T., R.F. Engle, and J.M. Wooldridge. 1988. A Capital Asset Pricing Model wit Time-Varying Coariances. Te Jornal of Political Economy 96:116-131. Brown, S.P., and M.K. Ycel. 008. Wat Dries Natral Gas Prices? Energy Jornal 9():45-60. Cang, C., M. McAleer, and R. Tanscat. 010. Analyzing and Forecasting Volatility Spilloers, Asymmetries and Hedging in Major Oil Markets. Energy Economics 3(6):1445-1455. Elder, J., and A. Serletis. 010. Oil Price Uncertainty. Jornal of Money, Credit & Banking (Wiley-Blackwell) 4(6):1137-1159. Engle, R.F. 198. Atoregressie Conditional Heteroscedasticity wit Estimates of te Variance of United Kingdom Inflation. Econometrica 50(4):987-1007. Engle, R. F., & Kroner, K. F. (1995). Mltiariate Simltaneos Generalized ARCH. Econometric Teory, 11(1):1-150. Ewing, B.T., F. Malik, and O. Ozfidan. 00. Volatility Transmission in te Oil and Natral Gas Markets. Energy Economics 4(6):55-538. Frenc, K.R., G.W. Scwert, and R.F. Stambag. 1987. Expected Stock Retrns and Volatility. Jornal of Financial Economics 19:3-9. Geman, H., and S. Oana. 009. Forward Cres, Scarcity and Price Volatility in Oil and Natral Gas Markets. Energy Economics 31(4):576-585. 4
Goodwin, B.K., and R. Scnepf. 000. Determinants of Endogenos Price Risk in Corn and Weat Ftres Markets. Te Jornal of Ftres Markets 0:753-774. Hamilton, J.D. 1983. Oil and te Macroeconomy since World War II. Jornal of Political Economy 91():8-48. Hamilton, J.D., and A.Herrera. 004. Oil Socks and Aggregate Macroeconomic Beaior: Te Role of Monetary Policy. Jornal of Money, Credit & Banking (Oio State Uniersity Press) 36():65-86. Hammode, S., H. Li, and B. Jeon. 003. Casality and Volatility Spilloers among Petrolem Prices of WTI, Gasoline and Heating Oil in Different Locations. Nort American Jornal of Economics & Finance 14(1):89-114. Hartley, P.R., K.B. Medlock III, and J.E. 008. Rostal. Te Relationsip of Natral Gas to Oil Prices. Energy Jornal 9(3):47-65. Kilian, L. 008. Exogenos Oil Spply Socks: How Big Are Tey and How Mc Do Tey Matter for te U.S. Economy? Reiew of Economics and Statistics 90(): 16-40. Lanman, Scott, and Ric Miller. 008. Bernanke May Rn Low on Ammnition for Loans, Rates. Bloomberg. Bloomberg, 17 Mar 008. Web. <ttp://www.bloomberg.com/apps/news?pid=newsarcie&sid=ao1ufsecliu& refer= ome>. Lee, T.K., and J. Zyren. 007. Volatility Relationsip between Crde Oil and Petrolem Prodcts. Atlantic Economic Jornal 35(1):97-11. Lin, W., and C. Dan. 007. Oil Conenience Yields Estimated nder Demand/Spply Sock. Reiew of Qantitatie Finance & Acconting 8():03-5. M, X. 007. Weater, Storage, and Natral Gas Price Dynamics: Fndamentals and Volatility. Energy Economics 9(1):46-63. Myers, R.J. 1991. Estimating Time-Varying Optimal Hedge Ratios on Ftres Markets. Te Jornal of Ftres Markets 11:39-53. Myers, R.J., and S.D. Hanson. 1993. Pricing Commodity Options wen te Underlying Ftres Price Exibits Time-Varying Volatility. American Jornal of Agricltral Economics 75:11-130. Olowe, R.A. 010. Oil Price Volatility, Global Financial Crisis and te Mont-of-te- Year Effect. International Jornal of Bsiness & Management 5(11):156-170. 5
Pindyck, R.S. 001. Te Dynamics of Commodity Spot and Ftres Markets: A Primer. Energy Jornal :1-9. Raman, S., and A. Serletis. 010. Te Asymmetric Effects of Oil Price and Monetary Policy Socks: A Nonlinear VAR Approac. Energy Economics 3(6):1460-1466. Regnier, E. 007. Oil and Energy Price Volatility. Energy Economics 9(3):405-47. Serletis, A., and J. Herbert. 1999. Te Message in Nort American Energy Prices. Energy Economics 1(5):471-483. Serletis, A., and R. Rangel-Riz. 004. Testing for Common Featres in Nort American Energy Markets. Energy Economics 6(3):401-414. Senaga, H., A. Smit, and J. Williams. 008. Volatility Dynamics of NYMEX Natral Gas Ftres Prices. J. Ftres Mkts. 8:438 463. Yang, S.R., and B.W. Brorsen. 1993. Nonlinear Dynamics of Daily Ftres Prices: Conditional Heteroskedasticity or Caos? Te Jornal of Ftres Markets 13(1):175-191. Zagaglia, P. 010. Macroeconomic Factors and Oil Ftres Prices: A Data-Ric Model. Energy Economics 3(): 409-417. 6
Table 1. Smmary Statistics Variable Mean Standard Deiation Minimm Maximm Crde Oil Retrn 0.040.337-16.544 18.444 Heating Oil Retrn 0.038.7-13.965 10.97 Natral Gas Retrn 0.01 3.54-1.617 3.586 % Δ Consmer Price Index 0.175 0.355-1.700 1.100 % Δ Indstrial Prodction Index 0.110 0.616 -.800 1.700 Treasry Bond Spread 1.019 0.939-0.410.830 % Δ Crde Oil Inentories 0.0 0.413-1.430 1.513 % Δ Heating Oil inentories -0.017 1.841-8.440 5.730 % Δ Natral Gas Inentories 0.155 4.771-17.357 1.346 Notes. Sample period is 0/01/1994-0/04/011 and total nmber of obserations is 476. Retrns are calclated as r t =100x(ln F t - ln F t-1 ), were F t is daily settlement price of te ftres contract on day t. Treasry bond spread is calclated as te difference between te 10-year and -year Treasry constant matrity rate and stated in percent. All economic ariables are interpolated ia a step fnction to obtain daily series to se wit te daily retrn data. Table. Agmented Dickey-Fller Unit Root Tests Variable τ p-ale Ftres Prices F CL -0.91 0.7851 F HO -0.6 0.864 F NG -.45 0.185 Log of Ftres Prices ln F CL -1.04 0.7414 ln F HO -0.70 0.8460 ln F NG -.11 0.418 Ftres Retrns r CL -48.75 <0.0001 r HO -47.38 <0.0001 r NG -47.31 <0.0001 Notes. Te τ statistics and teir p-ales are presented for single-mean Agmented Dickey-Fller nit root test wit one lag. CL, HO, and NG refer to crde oil, eating oil, and natral gas, respectiely. Ftres retrns are calclated as r t =100x(ln F t - ln F t-1 ). 7
Table 3. GARCH-BEKK Reslts Mean Eq. CL HO NG Constant 0.07 0.080 0.061 (0.000) (0.000) (0.181) R t-1 0.037 0.03-0.030 (0.000) (0.000) (0.0) R t- -0.044-0.040-0.0 (0.000) (0.000) (0.043) R t-3-0.031-0.031-0.004 (0.000) (0.000) (0.703) R t-4-0.010-0.001 0.015 (0.10) (0.901) (0.17) R t-5-0.001 0.001-0.04 (0.858) (0.830) (0.056) Variance Eq. Var (CL) Var(HO) Var (NG) Co (CL,HO) Co (CL,NG) Co (HO,NG) Constant 10.417.913.70 5.416-5.187 -.804 (0.000) (0.017) (0.000) (0.000) (0.000) (0.000), 0.00 0.007 0.003-0.01-0.008 0.005 (0.003) (0.038) (0.549) (0.000) (0.41) (0.50), 0.00 0.074 0.00 0.014 0.00 0.013 (0.333) (0.000) (0.656) (0.086) (0.414) (0.375), 0.000 0.000 0.045 0.000-0.003-0.00 (0.543) (0.533) (0.000) (0.56) (0.18) (0.06),, 0.014-0.045-0.005 0.035 0.004-0.019 (0.010) (0.001) (0.584) (0.000) (0.551) (0.4),, -0.003 0.00-0.04 0.000 0.031-0.017 (0.30) (0.8) (0.38) (0.441) (0.000) (0.000),, -0.001-0.006 0.00-0.004 0.010 0.057 (0.365) (0.6) (0.373) (0.44) (0.068) (0.000), 0.908 0.001 0.000 0.033 0.016 0.001 (0.000) (0.067) (0.811) (0.000) (0.63) (0.63), 0.000 0.816 0.064-0.05-0.007 0.8 (0.53) (0.000) (0.003) (0.019) (0.037) (0.000), 0.004 0.00 0.134 0.003 0.04 0.018 (0.000) (0.004) (0.000) (0.000) (0.000) (0.000), -0.05 0.06 0.008 0.860 0.41 0.04 (0.04) (0.000) (0.611) (0.000) (0.000) (0.436), 0.17 0.003 0.01 0.049 0.350 0.013 (0.000) (0.005) (0.631) (0.000) (0.000) (0.001), -0.004 0.088 0.185 0.059 0.007 0.343 (0.063) (0.000) (0.000) (0.000) (0.145) (0.000) 8
Variance Eq. Var (CL) Var(HO) Var (NG) Co (CL,HO), 9 Co (CL,NG) Co (HO,NG) 0.140 0.03 0.013 0.067-0.04-0.00 (0.000) (0.033) (0.619) (0.00) (0.34) (0.316), 0.17 0.034 0.017 0.066-0.046-0.04 (0.000) (0.090) (0.604) (0.013) (0.99) (0.87), 0.001 0.001 0.006 0.001 0.003 0.00 (0.348) (0.41) (0.518) (0.373) (0.35) (0.340),, -0.67-0.067-0.09-0.133 0.089 0.044 (0.000) (0.047) (0.601) (0.003) (0.97) (0.85),, 0.06 0.011-0.017 0.017 0.05 0.011 (0.064) (0.130) (0.439) (0.090) (0.68) (0.37),, -0.05-0.011 0.00-0.017-0.03-0.011 (0.087) (0.179) (0.37) (0.11) (0.310) (0.431) CPI 1.044 4.105 0.008-1.163-0.056 0.183 (0.039) (0.001) (0.879) (0.045) (0.771) (0.750) CPI*I - 8.03 1.077.388 -.097-3.940 1.408 (0.00) (0.411) (0.186) (0.50) (0.011) (0.34) IP 0.097 0.660 0.000 0.04 0.000 0.001 (0.41) (0.04) (0.997) (0.181) (0.994) (0.994) IP*I - 0.580 3.7 0.393-1.148-0.399 1.134 (0.15) (0.048) (0.68) (0.086) (0.166) (0.018) TSPRD 0.130 0.061 0.06 0.088-0.089-0.061 (0.01) (0.096) (0.015) (0010) (0.001) (0.007) INV CL 0.06 0.404 0.489 0.096 0.101 0.443 (0.636) (0.18) (0.07) (0.431) (0.448) (0.00) INV CL*I - 0.5 1.501 0.043-0.646-0.119 0.54 (0.45) (0.175) (0.701) (0.199) (0.453) (0.466) INV HO 0.038 0.043 0.001-0.016-0.003 0.006 (0.004) (0.073) (0.615) (0.188) (0.345) (0.331) INV HO*I - 0.031 0.116 0.045-0.06-0.01 0.069 (0.070) (0.070) (0.00) (0.35) (0.8) (0.004) INV NG 0.001 0.001 0.000 0.000 0.000 0.000 (0.01) (0.61) (0.874) (0.965) (0.880) (0.844) INV NG*I - 0.015 0.011 0.000-0.001 0.000 0.00 (0.015) (0.1) (0.801) (0.747) (0.660) (0.606) ASNFC 0.091.666 0.635 0.387-0.198-1.96 (0.589) (0.018) (0.87) (0.49) (0.419) (0.091) OPEC 1.180 0.560 0.03 0.733-0.15-0.11 (0.05) (0.074) (0.691) (0.01) (0.391) (0.46) SEP11 5.06 0.78 0.158 1.179 0.889 0.10 (0.000) (0.34) (0.157) (0.014) (0.008) (0.094) IRQINV 1.76 0.49 1.638 0.43 0.175 0.575 (0.054) (0.63) (0.106) (0.33) (0.701) (0.38) KATRN 4.880 1.381 8.943 -.556 6.593-3.490 (0.005) (0.133) (0.009) (0.00) (0.000) (0.010) LEHMN 5.116 0.051 8.93-0.470-6.394 0.634 (0 007) (0 883) (0 018) (0 79) (0 000) (0 778)
Variance Eq. Var (CL) Var(HO) Var (NG) Co (CL,HO) Co (CL,NG) Co (HO,NG) Mon 0.334 1.168 3.04 0.495 0.75 1.861 (0.093) (0.015) (0.000) (0.043) (0.040) (0.000) Te 0.008 0.074 0.30-0.006 0.007-0.130 (0.60) (0.471) (0.077) (0.899) (0.98) (0.169) Wed 0.089 0.306 0.47-0.11 0.095-0.60 (0.53) (0.099) (0.059) (0.00) (0.31) (0.013) T 0.451 0.008 1.581-0.059 0.86-0.108 (0.067) (0.845) (0.000) (0.699) (0.000) (0.707) Jan 0.50 0.583 0.05-0.31 0.064-0.11 (0.93) (0.310) (0.664) (0.369) (0.488) (0.437) Feb 1.989 0.16 0.43-0.510-0.649 0.196 (0.010) (0.568) (0.9) (0.71) (0.01) (0.334) Mar 3.706 0.535 1.311-1.384 -.165 0.838 (0.000) (0.73) (0.007) (0.08) (0.000) (0.053) Apr 4.75 0.66 3.936-1.656-4.17 1.569 (0.000) (0.63) (0.000) (0.04) (0.000) (0.034) May 1.8 0.449 1.705 0.40 1.668-0.6 (0.011) (0.09) (0.009) (0.395) (0.000) (0.19) Jn 3.341 0.667 1.618-1.476 -.76 1.037 (0.000) (0.354) (0.04) (0.069) (0.000) (0.146) Jl.36 0.06 0.347-0.18-0.896 0.081 (0.009) (0.818) (0.148) (0.714) (0.011) (0.731) Ag 1.61 0.510 0.033 0.89-0.0-0.19 (0.030) (0.36) (0.66) (0.073) (0.339) (0.37) Sep 3.3.999 0.077 3.110 0.498 0.480 (0.014) (0.018) (0.616) (0.00) (0.305) (0.300) Oct 3.009 1.533 0.135.135 0.63 0.456 (0.007) (0.16) (0.95) (0.007) (0.057) (0.076) No 1.853.3 0.09 1.975-0.596-0.680 (0.030) (0.03) (0.89) (0.004) (0.047) (0.058) LLF -5619.9 LR 883.793 (0.000) Lyng-Box Q 5.015 4.346 3.71 (0.096) (0.370) (0.787) Notes. Te transformed coefficients on eac term in te ariance and coariance eqations and teir p- ales are presented. LLF refers to loglikeliood fnction ale. Likeliood ratio (LR) test statistics and its p-ale for te nll ypotesis of no exogenos ariables in ariance eqations are gien. Lyng-Box Q statistics and teir p-ales for te test of independence of te model residals are presented. 30