The Influence of Oil Prices on Indonesia’s Exchange Rate	305

Various studies show difference in the behaviours of macroeconomic factors at different exchange rate regimes (Mundell, 1995; Rolnick and Weber, 1997; Yeyati and Sturzenegger, 2003; Husain, Mody, and Rogoff, 2005). Rolnick and Weber (1997) show that output growth is higher under fiat standards than under commodity standards. In their study of the association between de facto exchange rate regimes and economic growth over the post–Bretton Woods period (1974–2000), Yeyati and Sturzenegger (2003) find that, in developing countries, less flexible exchange rate regimes are associated with slower growth while more flexible regimes are associated with greater output volatility. The authors find no such links for industrial countries (see also Mundell, 1995). Husain, Mody, and Rogoff (2005) evaluate regime performance in terms of inflation, growth, and crisis outcomes between developing, emerging, and advanced economies. They find evidence that, for developing countries, fixed exchange rate regimes lower inflation and more flexible regimes are associated with higher inflation, but with no evident gains in growth. They find similar evidence for emerging countries, but with small differences (and not always significant). On the other hand, more flexible regimes in advanced countries are associated with lower inflation and higher growth.

The question we ask in this paper, is whether the reaction of the exchange rate to oil prices differs between the managed float and float regimes. Evidence documented in this literature, particularly those from the work of Husain et al. (2005) imply that we expect to see differences in the reaction of the exchange rate to oil price changes between the managed float and float regimes. Husain et al. (2005) show that, for developing nations, inflation is lower in fixed exchange rate regimes compared to more flexible regimes. Hence, for Indonesia, we expect that, under the managed float regime, exchange rate management will be more in tune with changes in oil prices than under the float regime. In other words, the effect of oil prices will be lower under managed float regime than the float regime.

Interestingly, current evidence on the reaction of the exchange rate to oil price changes covers managed float and float regimes with no distinction between the regimes. Further evidence suggests no short-term link between the exchange rate and oil prices. Narayan (2013) finds, under a predictive modelling framework, that Indonesia’s exchange rate, in nominal terms, is unrelated to oil prices in both the in-sample and out-of-sample forecasting exercises. Moreover, the author shows that the exchange rate is significantly related to the oil prices of other nations, such as Vietnam, Bangladesh, Cambodia, and Hong Kong. Recently, Narayan and Sahminan (2018) and Narayan et al. (2019) have re-examined the exchange rate model in real terms, where the main focus is on the implications of cryptocurrency and fintech, respectively. More importantly, these studies also account for the effects of oil prices but find prices to be insignificantly related to the RER.

The disconnect between the exchange rate and oil prices that we note in the literature is a feature of short-term models. The aim of this paper is to test whether this disconnect between oil prices and the exchange rate is just a short- term phenomenon or also a long-term phenomenon. One likely explanation for this disconnect is that studies have not accounted for changes in the exchange rate regimes (Narayan and Sahminan, 2018). Therefore, we use an exchange rate model similar to that proposed for Indonesia by Narayan et al. (2018) and

306Bulletin of Monetary Economics and Banking, Volume 21, Number 3, January 2019

Narayan and Sahminan (2018). One difference is that we cover a longer sample period, which allows us to examine the exchange rate–oil price nexus under the managed exchange rate regime prior to August 1997 and a floating exchange rate regime (post–August 1997) for Indonesia. We use a battery of cointegration tests to examine any possible long-run relationship between the exchange rate and oil prices.

The present study is, to the best our knowledge, the first to examine the oil price-exchange rate nexus under different exchange rate regimes (i.e. floating and managed floating systems). Almost every other study on the oil price–exchange rate relation focuses on short-term linkages, with none focused on the effects of different exchange rate regimes (Amano and van Norden, 1998; Camarero and Tamarit, 2002; Chen and Chen, 2007; Lizardo and Mollick, 2010; Basher et al., 2012; Narayan, 2013; Narayan and Sahminan, 2018; Narayan et al. 2018).

II.EXCHANGE RATE REGIMES AND OIL EXPORTS AND IMPORTS IN INDONESIA: 1986–2018

A. Exchange Rate Regimes

Over our study period, 1986–2018, Indonesia moved to a new exchange rate regime once. On 14 August 1997, the move to a float regime from a managed float regime was undertaken in the wake of the Asian financial crisis, to prevent further depletion of foreign exchange reserves. The Asian financial crisis resulted in significant capital flight and increased speculation activities against the rupiah that weakened the rupiah exchange rate. This condition was exacerbated by social unrest and political instability in the country. The exchange rate crisis accompanied by social turmoil in the country resulted in hyperinflation and deep economic contraction in 1998.

The subsequent economic recovery that came with a more stable social, economic, and political environment saw the rupiah gaining ground by 2003. To date, the country continues to follow a floating exchange rate regime, where Bank Indonesia implements exchange rate stabilization measures in line with the currency’s fundamental value. At the same time, Bank Indonesia strives to maintain market mechanisms backed by financial market–deepening efforts.

B. Exports and Imports of Crude Oil and Partly Refined Petroleum

From 1989 to 2017, crude oil with partly refined petroleum exports declined by 65%, which was less than the drop of 85%in exports of partly refined petroleum only. On the other hand, the import of partly refined petroleum increased more than the import of crude oil and partly refined petroleum (261%). This means that, since the early 2000s, Indonesia has become increasing reliant on imported partly refined petroleum products. During this period, crude oil and partly refined petroleum (HS 2709) exports averaged 74,076 tonnes per day while imports averaged 34,900 tonnes per day (Figure 1). Over the same period, excluding crude oil, partly refined petroleum (HS 2710) exports and imports averaged 15,763 and 43,010 tonnes per day, respectively.

The Influence of Oil Prices on Indonesia’s Exchange Rate	307

According to the United Nation’s Comtrade Database, Indonesia was a net exporter of crude oil and partly refined petroleum up until 2012. From 2013 to 2017, net imports of crude oil plus refined petroleum averaged 8,735 tonnes per day. Prior to this period (1989–2012), net exports of crude oil plus refined petroleum averaged 49,158 tonnes per day.

However, for partly refined petroleum, excluding crude oil, Indonesia became a net importer much earlier, in 1996 (Figure 1) and its net imports (imports minus exports) of refined petroleum averaged 42,037 tonnes per day from 1996 to 2017.

Figure 1. Indonesia’s Exports and Imports of Crude and Refined Petroleum

(in Million Tons)

This figure depicts exports and imports of ‘Petroleum oils, oils from bituminous minerals, crude’ (HS 2709) and exports and imports of ‘Oils petroleum, bituminous, distillates, except crude’ (HS 2710). We also provide a balancing figure that is derived after subtracting imports from exports. We refer to this balancing figure net exports (net imports) for HS 2709 and HS 2710, if it takes a positive (negative) value.

50

40

30

20

10

-
(10)					Balance HS 2709									Balance HS 2710
(20)
					Exports HS 2709 (Crude)									Exports HS 2710 (Except Crude)
(30)					Imports HS 2709 (Crude)									Imports HS 2710 (Except Crude)
1989				1990		1991	1992	1993	1994	1995	1996	1997	1998	1999	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017

Source : UN Comtrade

III. THEORY AND EMPIRICS

Theoretically, higher oil prices should lead to the transfer of wealth between the exporter and importer of oil (Golub, 1983; Krugman, 1983; Corden, 1984; De Grauwe, 1996). Higher (lower) prices could see appreciation (depreciation) of the exporter currency against the importer currency. However, since the US dollar is the major invoicing and settlement currency in the international market, theoretically, higher (lower) energy prices will increase (reduce) demand for the US dollar (Zhang et al., 2008). In return, increased (reduced) demand for the US dollar should lead to depreciation (appreciation) of the currency of (non-US) importers of energy sources against US currency. Further, if higher prices of crude oil occur simultaneously with higher demand for oil by a non-US importer, the effect could be a much greater depreciation in the non-US importer currency against the US dollar.

308Bulletin of Monetary Economics and Banking, Volume 21, Number 3, January 2019

This depreciating effect of higher oil prices for the currencies of other industrialized nations against the US dollar has been noted in various studies (Amano and van Norden, 1998; Camarero and Tamarit, 2002; Chen and Chen, 2007; Lizardo and Mollick, 2010; Basher et al., 2012). Pershin et al. (2016) find that the net oil importing sub-Saharan countries of Botswana, Kenya, and Tanzania behaved differently before and after the oil price shock of July 2008. Ghosh (2011) finds that higher oil prices led to depreciation of the Indian rupee vis-á-vis the US dollar from 2 July 2007 to 28 November 2008. However, Narayan et al. (2008), in a study of Fiji, a net importer of petroleum, show that higher oil prices led to short-term appreciation of the Fiji dollar vis-á-vis the US dollar from 2000 to 2006. Narayan (2013) finds mixed results for selected Asian nations, including Indonesia. The author finds that oil prices are a good in-sample predictor of the nations’ exchange rate against the US dollar, although, for some of these countries, a higher oil price was predicted to appreciate the local currency against the US dollar. Narayan (2012) finds that oil prices are not a (short-term) predictor of Indonesia’s rupiah against the US dollar. These mixed effects in non-OECD countries could be related to changes in the exchange rate regimes over time or the exchange rate regime been used at the point in time (see discussion in Section 1).

Following uncovered real interest rate parity and the well-known Balassa– Samuelson model, we also consider productivity, inflation, and interest rate differentials as theoretically important determinants of the RER. With the exception of Chen and Chen (2007), Narayan et al. (2018), Narayan and Sahminan (2018), the focus has been primarily on the exchange rate and oil prices. We follow the broader exchange rate literature to augment this model to one that includes productivity and real interest rate differentials.

We differ from the literature in that we compare the linkage between the exchange rate and oil prices under different exchange rate regimes, in particular, managed float and float regimes. The motivation is obvious: the relation between exchange rate and oil prices is likely to be dependent on the exchange rate regime.

IV. DATA

Due to data limitations, a variety of frequencies and data samples were used to arrive at robust findings. The empirical analyses are conducted over three frequencie s: daily, monthly and annual. The real exchange rate (RER) and West Texas Intermediate (WTI), which proxies for oil prices, are our key variables. Inflation, the interest rate, and productivity differentials are available only at annual and monthly frequencies. Our daily models are in nominal terms, whereas the monthly and annual models are in real terms. The time period varies by data frequency. Daily data cover the period from 9 November 1991 to 26 November 2018, monthly data span the period from January 1986 to April 2018, and annual data cover the period from 1991 to 2017. These data sets cover periods during which Indonesia was a net importer of partly refined petroleum (1997 onwards) and of crude oil and partly refined petroleum (2013 onwards)

On the basis of data availability, daily and monthly data were examined for three subsamples: the full sample, the managed float sample (prior to 14 August 1997), and the float sample (14 August 1997 onwards). An important point is that

310Bulletin of Monetary Economics and Banking, Volume 21, Number 3, January 2019

Table 2.

Descriptive Statistics (Continued)

	Panel C: Annual data
Variables	Definition	Calculations	Source
WTI	Crude Oil Prices: West	USD per barrel	CEIC
	Texas Intermediate
RER	Real exchange rate,		Nominal exchange rate is
	expressed as the US dollar		sourced from CEIC; RER
	in terms of Rupiah. Increase		is calculated by the author.
	in the RER indicates
	depreciation of the Rupiah
	against the US dollar and
	vice versa.
DY	Difference of the	DY= YIndonesia-YUS, where	Indonesia and US RGDP
	productivity (Y) between	YIndonesia= Log(RGDPIndonesia)-	(USDb) and Employment
	the US and Indonesia	Log(EmploymentIndonesia)	(no. of person) data – CEIC;
		and	DY – author’s calculations

YUS= Log(RGDPUS)-

Log(EmploymentUS)

Descriptive statistics are presented in Table 3 from the daily and monthly series, we note that, on average, the rupiah is weaker against the US dollar in the float regime compared to the managed float period. The managed float regime is accompanied by a crawling band, which explains why the volatility, measured by the coefficient of variation, during this period is lower than that in the float regime.

Table 3.

Descriptive Statistics

This table presents descriptive statistics for variables in daily form: NER and WTI; Monthly: RER, WTI, and RIR; and Annual: RER,

WTI, RIR, and DY. The variables are defined in Table 2. Note the definition of the Rupiah-US exchange rate: aUS/Rupiah; b Rupiah/ US. The descriptive statistics are for the full sample, floating and managed-floating period. The columns entitled RERb and RER have their mean, maximum, and minimum, and std. dev. multiplied by 1000.

	Daily			Monthly			Annual
	9 Nov 1991-		Jan 1986-April 2018				1991-2017
Full sample	26 Nov 2018
	NERa	WTI	RERb	WTI	RIR	RERa	WTI	RIR	DY
Mean	8380	48.66	0.046	43.34	2.22	11029	46.83	1.75	-1.65
CV	0.45	0.62	0.23	0.68	2.72	0.27	0.63	2.35	-0.18
Maximum	16650	145.31	0.075	133.88	42.56	21066	99.67	11.88	-1.17
Minimum	1980	0.00	0.014	11.35	-43.50	7999	14.42	-7.19	-1.91
Std. Dev.	3741	30.08	0.01	29.61	6.06	3009	29.42	4.11	0.30
Obs.	6911	6911	383	388	337	28	28	20	28

	The Influence of Oil Prices on Indonesia’s Exchange Rate									311
					Table 3.
			Descriptive Statistics (Continued)
		Daily			Monthly			Annual
	Managed-	29 Nov 1991 -		Jan 1986-Jul 1997
		31 July 1997
	floating
		NER	WTI	RER	WTI	RIR
	Mean	2192	19.54	0.056	19.41	3.97
	CV	0.06	0.12	0.090	0.18	0.97
	Maximum	2633	26.55	0.075	36.04	12.90
	Minimum	1980	13.89	0.051	11.58	-3.01
	Std. Dev.	131	2.41	0.005	3.56	3.84
	Obs.	1480	1480	139	139	91
		2 August 1997-		Aug 1997-April 2018				1998-2017
	Floating	26 Nov 2018
		PRICE	WTI	RER	WTI	RIR	RER	WTI	RIR_	DY
									DIF
	Mean	10067	57	0.040	56.70	1.58	12011	57.39	1.21	-1.83
	CV	0.21	0.52	0.200	0.52	4.17	0.25	0.50	2.83	-0.02
	Maximum	16650	145.31	0.055	133.88	42.56	21066	99.67	4.99	-1.75
	Minimum	2582	0	0.014	11.35	-43.50	8593	14.42	-7.19	-1.91
	Std. Dev.	2127	29	0.008	29.35	6.58	3037	28.58	3.43	0.04
	Obs.	5431	5431	244	249	246	20	20	19	20
	Advent of			Aug 2011-April 2018
	Bitcoin			RER	WTI	RIR
	Mean			0.046	72.79	1.99
	CV			0.090	0.34	0.84
	Maximum			0.055	106.57	5.68
	Minimum			0.038	30.32	-1.28
	Std. Dev.			0.004	25.06	1.67
	Obs.			76	76	76

Looking at the monthly RER series over the float regime and the Bitcoin period, we note that the Bitcoin period coincides with an average appreciation of the rupiah against the US dollar. Further, the RER is less volatile in the period Bitcoin was introduced than in the period prior to its introduction.

Oil prices are, on average, higher during the float period than in the managed float regime. In recent years (which marks the advent of Bitcoin), oil prices have reached new heights. Oil prices were most volatile during the float period compared to the managed float and recent years.

The other two determinants of the RER are the real interest rate differential (RIR) and the productivity differential (DY) between Indonesia and the United States. The RIR is the most volatile series of all the data. The series most volatile in the float period but, on average, highest in the managed float period. The DY values are best developed with annual data (see Table 2 for definition).

Next, we examine the time series properties of our data. All variables, except RIR, are expressed in logarithmic form. The unit root test is performed before

312Bulletin of Monetary Economics and Banking, Volume 21, Number 3, January 2019

conducting the cointegration tests. We use three cointegration tests, of which the Engle–Granger (1987; hereafter EG) test and the Johansen (1998, 1991, 1995) test can only be conducted for I(1) variables and the third test, the autoregressive distributed lag (ARDL; see Pesaran and Shin, 1995) approach to cointegration, uses both I(0) and I(1) variables but not I(2) variables. We use the conventional augmented Dickey–Fuller (ADF) test to evaluate the null of a unit root against the alternative of no unit root.5 This test is conducted on all variables across the full sample and various subsamples for each of the three frequencies.

The results are reported in Table 4. Note that, as highlighted in Section II, the timeline for each frequency is different, which explains why we obtain different results across frequencies. We find that the daily oil price (WTI) and the nominal exchange rate (NER) are I(1) or stationary in the first differenced form in the full sample and all the other subsamples, except for NER in the (free) float period. This means that we can apply the Engle–Granger and Johansen methods to all the samples except during the float period when using daily data. The ARDL method is applicable to the full sample and subsamples of the daily data.

The monthly WTI and RER are stationary in level form the full sample and managed float periods but nonstationary in level form in the float regime, suggesting the applicability of all three methods of cointegration in the latter regime but only the use of the ARDL method in the former regime. All annual series are stationary after being differenced only once, which indicates that all three cointegrating methods apply when annual data are used.

Table 4.

Unit Root Test Results

This table presents the ADF test results, the test statistic and the corresponding probability value (in parenthesis) for all the variables used by three different data frequencies and sample periods. Lag length(s) were selected automatically using Akaike Information Criteria. The null of unit root is tested against the alternative of no unit root. Finally, *, **, and *** denote statistical significance at the 10%, 5% and 1% levels, respectively.

	Frequency:		Daily	Monthly		Annual
		Nominal		Real			Real
		I(0)	I(1)	I(0)	I(1)	I(0)	I(1)
	Full Sample	9 Nov 1991-		Jan 1986-April 2018		1991-2017
		26 Nov 2018
	WTI	-1.551	-84.665***	-3.456**		-1.061	-4.710***
	ER	[0.508]	[0.000]	[0.011]		[0.716]	[0.001]
		-1.822	-10.867***	-4.220***		-2.266	-5.903***
	DY	[0.370]	[0.000]	[0.001]		[0.190]	[0.000]
						-1.596	-5.071***
	RIR			-5.189***		[0.471]	[0.000]
				[0.000]

5Narayan and Popp (2010) structural break test was also conducted on full sample data on exchange rate. Results from all frequencies, except daily frequency, are consistent with the reported results. For daily data, Narayan and Popp test suggests stationarity at level form with breaks in 2005:07 and 2008:07. We estimated the daily full sample ARDL model with levels of the NER and the structural breaks and found the findings to be no different from the ones explained in the paper. These results are available on request.

The Influence of Oil Prices on Indonesia’s Exchange Rate						313
			Table 4.
	Unit Root Test Results (Continued)
Frequency:		Daily	Monthly		Annual
	Nominal			Real		Real
	I(0)	I(1)	I(0)	I(1)	I(0)	I(1)
Managed-	29 Nov 1991 -		Jan 1986-Jul 1997
floating regime	31 July 1997
WTI	-2.813	-38.024***	-4.047***
ER	[0.057]	[0.000]	[0.002]
	1.518	-16.642***	-4.206***
RIR	[0.999]	[0.000]	[0.001]
			-1.171	-8.812***
			[0.684]	[0.000]
Floating regime	2 August 1997-		Aug 1997-April 2018		1998-2017
	26 Nov 2018
WTI	-1.790	-75.333***	-2.484	-10.530***	-1.388	-4.296***
ER	[0.386]	[0.000]	[0.121]	[0.000]	[0.568]	[0.003]
	-5.418***		-1.866	-11.890***	-2.649	-5.279***
RIR	[0.000]		[0.348]	[0.000]	[0.100]	[0.000]
			-2.001	-15.371***
			[0.286]	[0.000]
Advent of			Aug 2011-April 2018
Bitcoin
WTI			-1.458	-6.330
ER			[0.550]	[0.000]
			-2.117	-8.933
RIR			[0.239]	[0.000]
			-2.491	-7.303
			[0.122]	[0.000]

V. RESULTS

A. Cointegration Between the Rupiah–US Dollar Exchange Rate and the Oil Price (WTI) As noted above, three cointegration tests, namely, the Engle–Granger, Johansen, and the ARDL tests, were conducted across all three data frequencies. The cointegration test results are reported in Tables 5 to 8. Table 5 summarizes the daily and annual cointegration test results and Table 6 presents the monthly results. All three methods are captured here. Tables 7 and 8 provide details on the ARDL models adopted, with Table 7 covering the daily and annual frequencies and Table 8 covering the estimated monthly ARDL models.

Two out of three cointegration tests’ results signal the absence of any cointegrating relationship between daily WTI and NER values in the full sample or under the managed floating regime. For both daily frequency subsamples, the results from the ARDL model suggest the presence of a stable long-run relationship, but further investigation suggests that these models fail diagnostic tests (Table 7). Hence, we are unable to find a robust cointegration relationship between NER and WTI. Looking at the monthly data, cointegration of the WTI and RER series is unanimously supported by all three cointegration tests in the float regime. However, we could not establish a stable long run link between RER and

314Bulletin of Monetary Economics and Banking, Volume 21, Number 3, January 2019

WTI in the more recent period (August 2011 onwards) which marks the advent of the Bitcoin. The ARDL model proves to be more supportive of a stable long-run relation with annual data for the full sample and the sample covering the float regime. Next, we estimate the long-run elasticities for the monthly and annual

models.

Table 5.

Daily and Annual Cointegration Between WTI and Exchange Rate: Full Sample

and/or Managed-Floating or Floating Regimes

This table presents the daily and annual data model-based results from tests of cointegration between WTI and RER from three different approaches to cointegration: Engle-Granger, Johansen, and the Autoregressive Distributed Lag (ARDL). The ARDL models and their diagnostics are presented in Table 7. Due to data limitations, the daily models comprise of nominal (N) variables while the annual models include variables in real (R) terms. For the diagnostics on the ARDL model, see Table 6. Finally, *, **, and *** denote statistical significance at the 10%, 5% and 1% levels, respectively. #MacKinnon (1996) p-values. ~Automatic lags specification based on Schwarz Information Criterion. ##MacKinnon-Haug-Michelis (1999) p-values.

	Panel 1: Daily			Full Sample					Managed-float regime
			Model 1: WTI, NER						Model 1: WTI, NER
	Engle-Granger	Dependent	tau-stat.	Prob.# z-Stat.		Prob.#	max	tau-stat. Prob.#		z-Stat.	Prob.#	max
							lag~					lag~
		NER	-1.793	0.633	-5.698	0.677	34	0.193	0.992	0.384	0.992	23
	Johansen		Unrestricted Cointegration Rank Test: Trace and Maximum Eigenvalue
			Trace		Max-Eigen			Trace		Max-Eigen
		No. of CE(s)	Stat.	Prob.##	Stat.	Prob.##		Stat.	Prob.##	Stat.	Prob.##
		None	6.562	0.629	4.189	0.839		11.198	0.200	9.685	0.233
		At most 1	2.374	0.123	2.374	0.123		1.513	0.219	1.513	0.219

ARDL	F-Stat.			Prob. F (10,6846)			F-Stat.		Prob. F
									(17,1453)
	86.600***			0.000			4.474***		0.000
Panel 2: Annual		Full Sample: 1991 2017						Float regime: 1998 2017
		Model 1: WTI, RER						Model 1: WTI, RER
Engle-Granger	Dependent	tau-	Prob.#	z-Stat.	Prob.#	max	tau-	Prob.#	z-Stat.	Prob.#	max
		statistic				lag~	statistic				lag~
	RER	-2.298	0.393	-8.581	0.391	5	-3.884**	0.036	-33.92***	0.000	3
Johansen		Trace		Max-Eigen			Trace		Max-Eigen
	No. of CE(s)	Statistic	Prob.##	Statistic	Prob.##		Statistic	Prob.##	Statistic	Prob.##
	None	9.940	0.285	6.881	0.503		9.940***	0.000	39.063***	0.000
	At most 1	3.059*	0.080	3.059*	0.080		3.059**	0.030	4.700**	0.030
ARDL	F-statistic		Prob(F-statistic)				F-statistic		Prob(F-statistic)
	4.176**		0.028				3.275*		0.063

The Influence of Oil Prices on Indonesia’s Exchange Rate	315

Table 6.

Monthly and Annual Cointegration Results: With More Variables

This table presents the monthly test results the cointegrating link between WTI and RER (with or without additional theoretically motivated variables) from three different approaches to cointegration: Engle-Granger, Johansen, and the Autoregressive Distributed Lag (ARDL). The ARDL models and their diagnostics are presented in Table 8. CE stands for cointegrating equations. Finally, *, **, and *** denote statistical significance at 10%, 5% and 1% levels, respectively.

				Monthly data
		Model	Model 1: WTI, RER				Model 2: WTI, RER, RIR
	Methods	Sample			Sample: 2011M08 2017M11
	Engle-	Dep. Var.	tau-stat.	Prob.#	z-Stat.	Prob.#	tau-	Prob.# z-Stat. Prob.#
	Granger						stat.
		RER	-9.017	0.000	-78.53	0.000	-2.581	0.451	-9.85	0.592
	Johansen		Trace		Max-Eigen		Trace		Max-Eigen
		No. of	Stat.	Prob.##	Stat.	Prob.##	Stat.	Prob.##	Stat.	Prob.##
		CE(s)
		None	54.523	0.000	27.96	0.000	34.175	0.015	19.63	0.079
		At most 1	26.559	0.000	26.55	0.000	14.491	0.07	10.75	0.177
		At most 2					3.916	0.048	3.916	0.048
	ARDL	F-stat.	0.111				0.994
		Prob.	0.895				0.422
		(F-stat.)
		Model	Model 1: WTI, RER				Model 2: WTI, RER, RIR
		Sample	Sample:1997M08 2018M0				Sample:1997M08 2018M0
	Engle-Granger	Dep. Var.	tau-stat.	Prob.#	z-Stat.	Prob.#	tau-	Prob.#	z-Stat.	Prob.#
							stat.
		RER	-5.164	0.000	-41.67	0.000	-5.371	0.000	-42.81	0.001
	Johansen		Trace		Max-Eigen		Trace		Max-Eigen
		No. of	Stat.	Prob.##	Stat.	Prob.##	Stat.	Prob.##	Stat.	Prob.##
		CE(s)
		None	30.679	0.000	27.05	0.000	91.086	0.000	55.88	0.000
		At most 1	3.626	0.057	3.626	0.057	35.206	0.000	29.21	0.000
		At most 2					5.996	0.014	5.996	0.014
	ARDL	F-stat.		4.017				213.574
		Prob.		0.000				0.000
		(F-stat.)
		Annual (1991-2017); Model 2: WTI, RER, DY
	Engle-Granger	Dep. Var.	tau-stat.		Prob.#		z-stat.		Prob.#
		RER	-5.148		0.006		-57.98		0
	Johansen			Trace				Max-Eigen
		No. of	Stat.		Prob.		Stat.		Prob.
		CE(s)
		None	35.6		0.009		22.6		0.03
		At most 1	13		0.114		7.656		0.414
		At most 2	5.343		0.02		5.346		0.02
	ARDL	F-stat.				171.07
		Prob.				0
		(F-stat.)

The Influence of Oil Prices on Indonesia’s Exchange Rate	319

B. Long-run Relationship Between the RER and Oil Prices (WTI)

In the previous section, we established a long-run cointegrating relationship between monthly and annual RER and WTI values in the float regime. The relationships we focus on (the long-run impact of oil prices on the exchange rate) are dictated by the fact that oil prices and the exchange rate are cointegrated, that is, they share a stable long-run relationship. The long-run models are estimated following the often prescribed dynamic ordinary least squares (DOLS) and fully modified ordinary least squares (FMOLS) methods.

We could not test cointegration for monthly the RER and WTI values in the full sample and managed float regime, since these variables are stationary in level form. For these two samples, we estimate the short-term or first-differenced regression models using robust least squares.

The empirical results are reported in Table 9. Looking at the annual models, we derive the long-run effects of WTI on RER for two samples: the full sample (1991– 2007) and the floating regime sample (1998–2016). For both these samples, we find that WTI has a negative effect on the RER, although the effect of WTI is stronger in the float regime period than in the full sample. Since the annual RER is measured as foreign currency in terms of the Indonesian rupiah, the negative effect of WTI suggests that higher WTI values reduce the RER or lead to an appreciation of the rupiah against the US dollar. We note this result holds even with the inclusion of additional variables such as DY.

For the monthly RER and WTI, the short-run relationship is negative in the full sample and the managed float sample. The monthly RER data is measured as rupiah in US dollar terms, which means that higher WTI decreases the RER, leading to a depreciation of the rupiah against the US dollar. This is a short-term result that holds when we add a determinant (RIR) of RER. However, in the float regime, where we are able to estimate long-run relationships, we find that the link between monthly RER and WTI values is positive, which implies that higher WTI values increase the RER, leading to an appreciation of the rupiah against the US dollar. This positive result holds with other control factors, such as the RIR (see Table 9). Taken together, the monthly results suggest that, in the managed float period, when Indonesia was always a net exporter of crude oil and partly refined petroleum, the RER and WTI price behave in a theoretically inconsistent manner, at least in the short run. However, during the float regime, when Indonesia was still a net exporter of crude oil for the most part, the appreciating effect of oil prices in the long run aligns well with theory.