Crude facts

Last year’s news is this year’s reality.

Late in the summer of 2014, the daily price of crude oil began plunging on the futures markets, where people agree to buy and sell oil on a given day in the future, at a given price.  The headlines predicted ruin for oil exporters.

The predictions were premature, since daily prices – and on a speculative market, no less – are too volatile to affect most production and consumption. For example, extracting oil from Kazakhstan’s huge field in the north Caspian Sea, Kashagan, has taken more than seven years and cannot possibly be contingent on the price of West Texas Intermediate crude on July 1, 2008.  While some contracts do specify the daily oil price for daily purchases under a long-run contract, the design of the contract itself depends on long-run prices. 

Particularly convenient for decision-makers is the annual price of Brent crude – a global benchmark – on the spot market, which hosts current purchases and sales.  One can think of this price as the annual average of daily prices, which looks backward.  In 2014, when daily oil prices began falling in the summer, the annual average had fallen only 9% since 2013, from about $109 to $99 (Figure 2).     

Today, however, when most media have concluded that the oil price decline has ended and so is no longer news, the annual Brent price has fallen 43%, to about $56 (Figure 1).  This has led producers to delay planned expansions of output.  The president of Kazakhstan, Nursultan Nazarbaev, said two weeks ago that the fall in energy prices had reduced government revenues by 40%, with more severe consequences than those of the global recession of 2007-2009, reported a Kazakhstani business weekly newspaper, Panorama. 

Kazakhstan, of course, has been in tighter spots.  National real income per capita – that is, purchasing power for a typical Kazakhstani – roughly halved in the early years of the transition to markets, in the early 1990s; and famine and cannibalism were common in the Stalinist 1930s.  Nevertheless, today’s slowdown in Kazakhstan is for real.  Real income per capita may rise less than 2% this year, judging from estimates of the government’s statistical committee.  In the heyday of high oil prices, growth rates closer to 10% prevailed.

What does the future bode?  That depends on prior conditions.  If the status quo continues, then I (crudely) estimate that the 2016 spot price of Brent crude may be about $61-$62 per barrel, roughly 9% higher than today.  (In contrast, the government of Kazakhstan cut its forecast of oil prices from $80 to $50 for its 2015-2017 budgets, reported Interfax-Kazakhstan in January.)  But if 2016 repeats 2014 – say, with a Chinese slowdown that sparks short sales in fossil-fuel markets – then the Brent price may fall to $49-$50. Will tomorrow’s reality be yesterday’s news?  – Leon Taylor tayloralmaty@gmail.com

 

Figure 1:  Annual spot price of Brent crude, 1987-2015.  Data source:  United States Energy Information Administration.

Figure 2:  Annual change in the yearly spot price of Brent crude, 1988-2015.  Data source: United States Energy Information Administration.

References

Interfax-Kazakhstan.  Kazakhstan revises 2015-2017 budget to base it on $50 per barrel.  January 16, 2015.  www.interfax.kz

 Панорама.  Президент поручил Кабмину разработать антикризисный план.  [Panorama.   The President requested his ministers to develop an anti-crisis plan.]  October 23, 2015.

Statistical Committee of the Ministry of the National Economy of the government of Kazakhstan.  Various statistics.  www.stat.gov.kz

United States Energy Information Administration.  Oil prices.  www.eia.gov     

Notes

To forecast the 2016 oil price, I estimated annual prices for 1987-2015 as averages of the monthly prices.  I specified an autoregressive model, since this suited the slow decline in the autocorrelation function and the sharp decline in the partial autocorrelation function for the sample. (Autocorrelation functions measure the strength of the relationship between two oil prices that differ in time by a given interval.  Correlations that are closer to 1 in absolute value indicate a stronger relationship.  For example, the correlation between the current oil price and last year’s price is .9.  For the current price and the price of two years before, the correlation is weaker, .77.)    

The analysis has three steps:  Control for long-run factors in oil prices; estimate short-run shocks; and forecast the oil price for a particular shock.

Step 1.  The first difference of oil prices is the gap between the current price and the previous year’s price, or P(t) – P(t-1).  The first difference was stationary; that is, its basic characteristics didn’t change over time.  This simplifies the statistical analysis.  So I regressed the first difference on its first and second lags:

Dependent Variable: DPRICE
Method: Least Squares
Date: 10/14/15   Time: 19:00
Sample (adjusted): 1990 2015
Included observations: 26 after adjustments
Variable	Coefficient	Std. Error	t-Statistic	Prob.
C	2.167438	3.369370	0.643277	0.5264
DPRICE(-1)	-0.024780	0.258721	-0.095778	0.9245
DPRICE(-2)	-0.180607	0.262893	-0.687000	0.4990
R-squared	0.020113	Mean dependent var		1.461330
Adjusted R-squared	-0.065095	S.D. dependent var		15.29756
S.E. of regression	15.78761	Akaike info criterion		8.464495
Sum squared resid	5732.718	Schwarz criterion		8.609660
Log likelihood	-107.0384	Hannan-Quinn criter.		8.506297
F-statistic	0.236042	Durbin-Watson stat		1.609951
Prob(F-statistic)	0.791637

The R-squared and F statistics – which measure the model’s overall accuracy -- are disappointing.  But the point of the model was to control for long-run influences, which are represented by the lags, and thus to obtain measures of short-run shocks as residuals.  I didn’t use the model to forecast.  

Step 2.  I regressed the first difference of oil prices on the residuals (“moving averages”) obtained from the first model.  As independent variables, I used the current residual as well as its first and second lags.  For the forecasts, I set the current residual equal to that obtained in 2015 (the “status quo” scenario) and then to that obtained in 2014 (the “bad news” scenario).  Solving for the current price in each scenario gave me the forecast. 

For example, the model below estimated the first difference of the price in the status-quo scenario.  The independent variables are the current residual (SE1_SHOCK_T) obtained in the first model, followed by the first and second lags of the residual.  In the forecast, I set SE1_SHOCK_T equal to the 2015 residual in the first model; SE1_SHOCK_T_MINUS1, also equal to the 2015 residual; and SE1_SHOCK_T_MINUS2, equal to the 2014 residual.

Dependent Variable: DPRICE
Method: Least Squares
Date: 10/14/15   Time: 20:51
Sample (adjusted): 1992 2015
Included observations: 24 after adjustments
Variable	Coefficient	Std. Error	t-Statistic	Prob.
C	5.186500	0.365028	14.20850	0.0000
SE1_SHOCK_T	0.991668	0.005341	185.6717	0.0000
SE1_SHOCK_T_MINUS1	-0.024016	0.006844	-3.508999	0.0022
SE1_SHOCK_T_MINUS2	-0.185787	0.007047	-26.36357	0.0000
R-squared	0.999442	Mean dependent var		1.509809
Adjusted R-squared	0.999358	S.D. dependent var		15.89140
S.E. of regression	0.402549	Akaike info criterion		1.169014
Sum squared resid	3.240921	Schwarz criterion		1.365356
Log likelihood	-10.02817	Hannan-Quinn criter.		1.221104
F-statistic	11941.26	Durbin-Watson stat		1.798218
Prob(F-statistic)	0.000000

My predictions are:

(1)    Status quo scenario (assumes that the 2016 shock equals the 2015 shock).  My mean estimate is $61.44.  The 95% confidence interval is [$60.71, $62.17].

(2)   Bad news scenario (assumes that the 2016 shock equals the 2014 shock).  My mean estimate is $49.75.  The 95% confidence interval [$49.01, $50.49].

(3)   Good news scenario (assumes that the 2016 shock equals the 2013 shock).  My mean estimate is $61.99. The 95% confidence interval is [$61.25, $62.73].

My dataset follows.  Its small size is definitely a constraint.

Price	Year
18.52	1987
14.95	1988
18.25	1989
23.68	1990
20.01	1991
19.31	1992
17.04	1993
15.84	1994
17.04	1995
20.64	1996
19.12	1997
12.78	1998
17.85	1999
28.52	2000
24.45	2001
24.96	2002
28.88	2003
38.23	2004
54.42	2005
65.15	2006
72.47	2007
96.85	2008
61.49	2009
79.51	2010
111.26	2011
111.65	2012
108.64	2013
99.02	2014
56.25	2015