2.1 Historical Data Set

We use historical annual data for the real economic activity measure, i.e., a dry bulk cargo freight rate index, global production and real prices of the respective four metals as well as the real prices for cotton, barley and coffee. We use the U.S. all urban consumers price index to adjust prices and the freight rate index for inflation. Data descriptions and sources can be found in the online-appendix.

Employing long sample periods, partly going back to 1879 for copper (the freight rates index is only available since 1879), 1900 for nickel, 1925 for cobalt and 1955 for lithium, allows us to estimate the long-run relationships between the variables. This is important due to the long investment cycles in the industry. However, historical data can come with measurement problems. This is particularly a concern for the cobalt and lithium market. These commodities were not traded on public exchanges for a long time. Their value chain and pricing are more complex than for copper and nickel. We have ensured that the data is as consistent as possible over time. We have also checked the history of these markets for signs of structural changes, which may be a moderate issue for the cobalt, lithium and nickel markets. We attribute some of the relatively broad sets of admissible draws to some remaining measurement errors.

Moreover, we use historical data on cotton prices since 1879. Cotton is a major non-metal input for industrial production. Its market is liquid and well documented. At the same time, its production and consumption should be uncorrelated to the ones of the selected metals, except for movements due to aggregate demand shock, hitting all commodities at the same time). This is an important assumption for our identification scheme. See the online-appendix for plots of the time series.

2.2 Metals Consumption Scenarios

The IEA (2021b) provides metals consumption forecasts for the stated policy scenario that is based on the status quo in early 2021 and the net-zero emissions (NZE) scenario. Figure 1 shows historical production levels for copper, nickel, cobalt and lithium and the future consumption paths in the two scenarios.

View Full Size

Metals consumption in the IEA’s net-zero emissions scenario and the stated policy scenario.

Citation: IMF Working Papers 2021, 243; 10.5089/9781513599373.001.A001

• Download Figure
• Download figure as PowerPoint slide

The NZE scenario is based on the premise that global temperature increases can be limited to 1.5°C in 2050. It assumes that there are net-zero CO2 emissions in 2050, including the energy sector. It implies that renewable energies become the leading source of electricity worldwide before 2030. In the transportation sector, the scenario assumes that electricity will cover 60 percent of energy consumption in addition to the broad use of hydrogen for trucks and shipping. Battery demand is expected to increase from 0.16 TWh in 2020 to 14 TWh in 2050, with 86 percent of the stock of cars being powered by electricity. We concentrate on this scenario which is the most ambitious with the highest chance of limiting global warming to 1.5°C (IPPC 2021).

The total consumption of lithium and cobalt would rise more than twentyfold and sixfold, respectively, driven by clean energy demand in the NZE scenario. Copper and nickel would see twofold and fourfold increases of total consumption, respectively. The NZE scenario of the IEA also implies that the consumption of the respective metal grows at a high rate between now and 2030, as the switch from fossil fuels to renewable energies requires large initial investments, but slow down in the later part of the scenario horizon.

Metals consumption in the stated-policy scenario follows more or less an extended historical trend.

3.1 Identification

Without further information it is not possible to identify B_0^{-1} and thereby the structural form in (2). The literature has come up with different restrictions, for instance placed directly on B_0^{-1} to solve this identification problem. We apply conventional sign restrictions (e.g., Faust, 1998, Canova and Nicolo, 2002, and Uhlig, 2005) and zero impact restrictions on the elements in B_0^{-1} , i.e., we assume that the structural shocks have either a positive, negative or no instantaneous effect on the endogenous variables on impact. We base these impact restrictions on economic intuition as specified in Table 1. We also impose narrative sign restrictions, which we explain further below.

^{Table 1:}

Sign and zero restrictions on impact effects

^{Table 1:}

Sign and zero restrictions on impact effects

	Global economic activity	Global metal production	Real metal price	Real cotton price
Aggregate commodity demand shock	+	+	+	+
Metal supply shock	+	+	-	0
Metal-specific demand shock		+	+	0

View Table

^{Table 1:}

Sign and zero restrictions on impact effects

	Global economic activity	Global metal production	Real metal price	Real cotton price
Aggregate commodity demand shock	+	+	+	+
Metal supply shock	+	+	-	0
Metal-specific demand shock		+	+	0

We interpret the first shock as an aggregate commodity demand shock that is related to the global business cycle and affects demand for all commodities at the same time. A positive shock increases global economic activity, the global production of the respective metal, and the prices of both the respective metal and cotton on impact.³

We label the second shock as a metal supply shock, capturing for example, strikes, or other production outages, or the earlier than expected opening up of a major mine. A positive shock that increases global metal production is assumed to drive up global economic activity, but to decrease the real metal price on impact. As it is specific to metal supply, it should have no effect on the price of cotton on impact.

We interpret the third shock as a metal-specific demand shock. A positive shock increases the production and price of the respective metal, but unlike the aggregate commodity demand shock we assume that the metal-specific demand shock has zero effect on the cotton price on impact. We do not make an a-priori assumption on its effect on global economic activity, as the direction could go both ways: The energy transition could increase energy costs and thus dampen growth. However, it could also foster economic growth by lowering energy costs in the long run depending on technological change and policies.

We assume that the metal specific demand shock characterizes most closely the energy transition in our structural scenario analysis. It raises the demand for specific metals but not for other commodities. Note that this shock may also include anticipation shocks due to changes in expectations about metal-specific future demand and supply.⁴ This is important, because the energy transition may affect metal markets also through this anticipation channel.

It is important for the scenario analysis that the metal-specific demand shock resembles the energy transition as close as possible. Narrative sign restrictions (Antolín-Díaz and Rubio-Ramírez, 2018) help us to sharpen the identification of the different structural shocks and thus the distinction between them. These restrictions are imposed on the importance of specific shocks during specific historical episodes (see Table 2). We source the events of the narrative sign restriction shocks displayed in Table 2 from historical market accounts from USGS (2013).

^{Table 2:}

Narrative sign restrictions

Note: AD = Aggregate commodity demand shock, MS = Metal supply shock, MD = Metal-specific demand shock, REA = Real Economic Activity Index, largest = the contribution of the shock to the fluctuation of the respective variable in the specified year is larger than the contribution of any other type of shock.

^{Table 2:}

Narrative sign restrictions

Metal	Year	Shock	Variable	Sign	Contribution	Narrative
Cobalt	1930	AD	REA	-	largest	Great Depression
	1994	MS	Price	-		Zaire declares autonomy
	2009	AD	REA	-	largest	Great Recession
	2009	AD	Price	-	largest	Great Recession
	2017	MD	Price	+	largest	EV batteries demand
Copper	1930	AD	REA	-	largest	Great Depression
	1930	AD	Price	-	largest	Great Depression
	1966	MD	Price	+	largest	Vietnam War
	1967	MS	Production	-		Strike
	2009	AD	REA	-	largest	Great Recession
	2009	AD	Price	-	largest	Great Recession
Lithium	2009	AD	REA	-	largest	Great Recession
	2017	MD	Price	+	largest	EV batteries demand
Nickel	1930/31	AD	REA	-	largest	Great Depression
	1969	MS	Price	-	largest	Strike
	1988	MD	Price	+	largest	Stainless steel demand
	2009	AD	REA	-	largest	Great Recession
	2009	AD	Price	-	largest	Great Recession

Note: AD = Aggregate commodity demand shock, MS = Metal supply shock, MD = Metal-specific demand shock, REA = Real Economic Activity Index, largest = the contribution of the shock to the fluctuation of the respective variable in the specified year is larger than the contribution of any other type of shock.

View Table

^{Table 2:}

Narrative sign restrictions

Metal	Year	Shock	Variable	Sign	Contribution	Narrative
Cobalt	1930	AD	REA	-	largest	Great Depression
	1994	MS	Price	-		Zaire declares autonomy
	2009	AD	REA	-	largest	Great Recession
	2009	AD	Price	-	largest	Great Recession
	2017	MD	Price	+	largest	EV batteries demand
Copper	1930	AD	REA	-	largest	Great Depression
	1930	AD	Price	-	largest	Great Depression
	1966	MD	Price	+	largest	Vietnam War
	1967	MS	Production	-		Strike
	2009	AD	REA	-	largest	Great Recession
	2009	AD	Price	-	largest	Great Recession
Lithium	2009	AD	REA	-	largest	Great Recession
	2017	MD	Price	+	largest	EV batteries demand
Nickel	1930/31	AD	REA	-	largest	Great Depression
	1969	MS	Price	-	largest	Strike
	1988	MD	Price	+	largest	Stainless steel demand
	2009	AD	REA	-	largest	Great Recession
	2009	AD	Price	-	largest	Great Recession

Note: AD = Aggregate commodity demand shock, MS = Metal supply shock, MD = Metal-specific demand shock, REA = Real Economic Activity Index, largest = the contribution of the shock to the fluctuation of the respective variable in the specified year is larger than the contribution of any other type of shock.

Examples are the Great Depression or the Great Recession, for which we specify aggregate commodity demand shocks as the most important drivers of economic activity and the copper and nickel prices. These crisis episodes have hit commodity markets broadly and should not be mistaken as shocks specific to the energy transition metals.

Historical episodes that come potentially closer to the metal-specific demand shock resembling the energy transition are the unexpected increase in stainless steel demand in 1988 that pushed up nickel prices or the unexpected rise in electric vehicle batteries demand in 2017 driving up lithium and cobalt prices. In 2017 lithium prices more than doubled and cobalt prices increased by 70%. These price increases might have represented first expectations of a nascent energy transition. It is noteworthy that 2017 lithium production adjusted quite strongly to demand and increased by over 80% during the same year.

3.2 Computation of Supply Elasticities

We obtain supply elasticities using the estimated B_0^{-1} matrix of structural impact effects and the reduced-form parameters A_i. The responses of the n = 4 variables in y_t to the structural shocks ε_t can be traced over time via {\Theta }_h={\varphi }_h{B}_0^{-1} for h = 1,2,… where Θ_h is an (n x n) matrix of structural impulse responses for the horizon h and {\varphi }_h={\displaystyle \sum }_{j=1}^h{\varphi }_{h-j}{A}_j and ϕ₀ = I_n (Lütkepohl, 2005).

The impact supply elasticity η_S is calculated as the ratio of the metal production response to a metal-specific demand shock (MD) relative to the price response to the same shock written as η_S = (Θ₀)_{M D,Prod}/(Θ₀)_{M D,Price}.⁵

Demand shocks shift the metal demand curve along the metal supply curve and thereby trace out its shape which gives the supply elasticity. Elasticities over longer horizons are based on the cumulative output response and the cumulative price change response and calculated as

3.3 Structural scenario analysis

We conduct structural scenario analysis for the price of each metal following the framework of Antolín-Díaz et al. (2021). Our object of interest is a conditional forecast y_T+1,T+h over the next h = 20 years for the endogenous variables, where T denotes the year 2020. The conditional forecast restricts some of the variables in y_T+1,T+h and a subset of the future shocks ε_T+1,T+h, thereby linking the path of future variables directly to certain shocks. We briefly lay out the underlying intuition tailored to the metal consumption scenarios from the IEA (2021b).

We take the consumption scenarios for each metal as given, thus pre-specifying the future metal consumption in the conditional forecasts y_T+1,T+h. We set global consumption equal to global metal production in the IEA scenarios, assuming that there are no short-term changes in inventories. The future paths of global economic activity, the metal price and the cotton price are left unspecified. Concerning the paths of future shocks, we constrain the aggregate commodity demand shock, the metal supply shock and the residual shock to their unconditional distributions. The algorithm then finds a series of metal-specific demand shocks that incentivizes the metal production path needed for the energy transition. From these shocks we derive the implied price and revenue paths.

Compared to traditional conditional forecasts, this methodology has the advantage that it can attribute the future path of endogenous variables to the path of a specific structural shock. As we deem the energy transition as a scenario resulting a series of the metal-specific demand shocks, it is important to specify this directly and not attribute the energy transition to exogenous increases in metal supply or some combination of other shocks.

For example, in our case the classical reduced-form conditional forecasting question is “What is the likely path of the metal price, given that metal production has to increase by a certain amount due to the energy transition?” The answer is confounded by a lack of causal structure. Metals prices could be high, boosting supply to reach the scenario output. However, it could also be the opposite: supply shocks could drive upward the supply, driving prices lower.

Due to the structural scenario framework, we can handle this reverse causality in the scenario. We can ask the more precise question “What is the likely price path if metal-specific demand shocks due to the energy transition increase metal production as needed?” The structural scenario is hence a conditional forecast of the variables in our model that generates the scenario metal output path with the restriction that only the commodity-market specific demand shock series can deviate from its unconditional distribution. The metal production and consumption path of the respective metal is exogenously given.

In the case of no restrictions, the endogenous variables’ unconditional forecast for periods T + 1 to T + h is given by

where y_T+i,T+h = (y_T+1…y_T+h) and b_T+1,T+h represent the deterministic part of the forecast, which depends on past observables, the reduced-form VAR parameters A_i for i = 1,…,p and the deterministic part D_t. The matrix M represents the effects of the structural shocks on future values of the endogenous variables as a function of the structural parameters in B_i and the reduced-form parameters in A_i (see Antolín-Díaz et al., 2021 or Waggoner and Zha, 1999 for further details). The unconditional forecast is independent of the structural parameters. It is distributed according to y_T+1,T+h ~ N(b_T+1,T+h,M’M), where M’M depends only on the reduced-form parameters.

To answer the question how the prices of energy transition metals fare in a net-zero emission scenario, we perform a restricted forecast of the endogenous variables {\tilde{y}}_{T+1,T+h}, for which we place restrictions both on parts of the future observable variables and future shocks. Hence, the future observables are restricted as

where \overline{C} is a (k₀ x nh) pre-specified selection matrix, including k₀ restrictions. The (k₀ x 1) vector {\overline{f}}_{T+1,T+h} denotes the mean of the constrained endogenous variables and the (k₀ x k₀) matrix {\overline{\Omega }}_f denotes the covariance restrictions, i.e., the uncertainty around the restrictions on the observables. In our baseline case we restrict the path for metal output according to the IEA scenarios and set {\overline{\Omega }}_f=0_{k_0}, thus assuming no uncertainty around the scenarios.

Secondly, we restrict k_s elements of the future shocks via the (k_s x nh) selection matrix Ξ expressed as \mathrm{\Xi }{\overline{\epsilon }}_{T+1,T+h}\sim N(g_{T+1,T+h},{\Omega }_g). The (k_s x 1) vector g_T+1,T+h denotes the mean and Ω_g the covariance restrictions on the shocks in the conditional forecast. Under invertibility of the VAR, the restricted shocks can be related to restrictions on the observables starting from equation (4) for the restricted future observables {\tilde{y}}_{T+1,T+h} via

yielding

where \underset{¯}{C}=\mathrm{\Xi }{(M\prime )}^{-1} and {\underset{¯}{\mathrm{\Omega }}}_f={\mathrm{\Omega }}_g. We would like to explain a pre-specified path in metal output (one component of {\tilde{y}}_{T+1,T+h}) via the metal-specific demand shock. The other shocks should occur according to their unconditional distribution. In other words, we would like to restrict these non-driving shocks, while leaving the metal-specific demand shock unspecified. Thus, we impose \mathrm{\Xi }{\tilde{\epsilon }}_{T+1,T+h}\sim N(0_{k_s},I_{k_s}) such that equation (8) becomes

The restrictions in equations (5) and (9) can then be stacked according to

where \hat{C}\prime =[\overline{C}\prime ,\underset{¯}{C}\prime ] such that the upper part relates to the conditions on observables and the lower part to the conditions on the shocks.

Antolín-Díaz et al. (2021) show how to solve for the restricted forecast of the observables {\tilde{y}}_{T+1,T+h} such that the restrictions in equation (10) hold. In our baseline application we place k₀ = 20 restrictions on the observables, i.e., future metal output is constrained to the IEA’s scenario output in each of the forecasted h = 20 periods. Moreover, we place k_s = 3 ⋅ 20 = 60 restrictions on the non-driving shocks, i.e., all shocks, except the metal-specific demand shock, are restricted to their unconditional distributions for the forecast horizon. Thus, the total number of restrictions k is equal to nh, the length of {\tilde{y}}_{T+1,T+h}. For the case k = nh, Antolín-Díaz et al. (2021) show that there exists a unique solution of the restricted forecast.

3.4 Estimation and Inference

Estimation and inference are based on standard Bayesian techniques laid out in Waggoner and Zha (1999), Rubio-Ramirez et al. (2010), and Antolín-Díaz et al. (2021). We use a Minnesota-type prior with standard shrinkage parameters (see Giannone et al., 2015) in combination with a sum-of-coefficients prior (Doan et al., 1984) and a dummy-initial-observation prior (Sims, 1993) to estimate equation (1) and the conditional forecasts.⁶ Our prior specification assumes that metal production growth is independent and identically distributed, while the log of the real activity index and the logs of the price levels follow a random walk.

Identification via sign restrictions (with additional zero restrictions) does not yield point estimates but instead sets of possible parameter intervals for the different elements in B_0^{-1}. For each model we obtain a set of 1,000 admissible draws, where each draw consists of a conditional forecast, future shocks, and an associated B_0^{-1} matrix that satisfies the identifying restrictions. These draws are also used for inference, i.e., they yield an indication of the uncertainty around the point-wise median estimates. We follow Antolín-Díaz and Rubio-Ramírez, 2018 and Antolín-Díaz et al., 2021 and report point-wise median and percentiles of impulse responses for set-identified structural VAR models, as it is common in the literature.⁷ The 1,000 different draws allow for the construction of credible sets by estimating an elasticity for each draw and then calculating percentiles.

The aim is to draw from a joint posterior distribution of both the structural parameters and the conditional forecast

where y^T is the historical sample, B'₊ = [B'1…B'_pΓ] collects the structural VAR lag coefficients including the exogenous parts , IR(B₀,B₊) are the identification restrictions and R({\tilde{y}}_{T+1,T+h},B_0,B_{+}) the structural scenario restrictions. Note that the structural scenario restrictions depend on the structural VAR parameters via equation (10).

To draw from this distribution, we use the algorithm from Antolín-Díaz et al. (2021) that builds on Waggoner and Zha (1999). The algorithm uses a Gibbs sampler procedure that iterates between draws from the conditional distributions of the structural parameters and the conditional forecast.⁸

Hence, we pick a random draw of structural parameters out of 25,000 potential draws that relies both on the actual data and on a structural forecast. We use the structural parameters from this randomly picked draw to then draw the scenario paths of the two price series and the economic activity index for the structural scenario that fits the specified metal production path. The next 25,000 draws for structural parameters rely on the original data and the data from the just drawn structural scenario.

4.1 Price Elasticity of Metals Supply

Supply elasticities summarize how fast firms raise output in reaction to a price increase. The model allows us to estimate these elasticities at different horizons for each of the metals.

The elasticities are based on the impulse responses of metal production and prices to a metal-specific demand shock, as shown in Figure 2.⁹ The impulse response functions show a significant increase in the prices of the four metals as a response to the metal-specific demand (MD) shock. The impact of that shock on production is muted for all metals, except for lithium.

View Full Size

Impulse responses of metal production and price to a metal-specific demand (MD) shock, that increases the median metal price by 10 percent on impact, for copper, nickel, cobalt and lithium based on 1,000 draws showing the pointwise median (red solid line) with 68% point-wise credible sets (red dotted lines) among the full set of impulse response functions for each model. The impulse response functions are derived from four different VAR models, one for each metal.

Citation: IMF Working Papers 2021, 243; 10.5089/9781513599373.001.A001

• Download Figure
• Download figure as PowerPoint slide

Figure 3 shows the estimates of the supply elasticities for copper, nickel, cobalt and lithium. Supply is quite inelastic over the short term, as it can only be expanded through more recycling and higher utilization of existing mining capacity. A demand-induced positive price shock of respectively 10 percent increases the same-year output of copper by 3.5 percent, nickel by 7.1 percent, cobalt by 3.2 percent and lithium by 16.9 percent.¹⁰

View Full Size

Supply elasticities at annual horizons based on the metal-specific demand (MD) shock with 68% point-wise credible sets. Elasticities are calculated via equation (3).

Citation: IMF Working Papers 2021, 243; 10.5089/9781513599373.001.A001

• Download Figure
• Download figure as PowerPoint slide

In the long run, however, supply is more elastic. Firms build new mines, innovate in extraction technologies and conduct exploration. After 20 years, the same price shock raises output of copper by 7.7 percent, nickel by 13.0 percent, cobalt by 8.6 percent and lithium by 25.5 percent.

The supply elasticities for lithium are much larger than for the other three metals. This is in line with the different ways of producing the four metals. Copper, nickel, and cobalt are extracted in mines, which often require capital intensive investment and involve long lead times of 16 years on average from exploration to construction (IEA, 2021b). In contrast, lithium is often extracted from mineral springs and brine, where salty water is pumped from the deep earth. Lead times to open new production facilities are much shorter with up to 7 years. Other factors such as innovation in extraction technology, market concentration and regulations also influence supply elasticities.

4.2 Price Forecasts

Results in Figure 4 show that the four metals are potential bottlenecks for the energy transition. Prices of cobalt, lithium, and nickel would rise several hundred percent from annual average levels in 2020 in the net-zero emissions scenario. The copper price would be up by about 60 percent, as it would face more moderate consumption increases. Inflation adjusted prices of the four metals would reach peaks roughly similar to previous historical price peaks. However, prices would stay at these high levels for more than a decade, far longer than during previous peak periods. Real prices for all four metals would broadly stay in the 2020 annual average range in the stated policy scenario.

View Full Size

Price scenarios for the stated policy scenario and the net-zero emissions scenario with 40% point-wise credible sets.

Citation: IMF Working Papers 2021, 243; 10.5089/9781513599373.001.A001

• Download Figure
• Download figure as PowerPoint slide

Prices peak mostly around 2030 for two reasons: First, the steep rises in demand are front-loaded in the net-zero emissions scenario. In contrast to fossil fuels-based energy production, which needs a flow of fossil fuels, renewable energy production only uses metals upfront for the construction of wind-turbines or batteries, for example. Second, the initial price boom induces a supply reaction, which reduces market tightness after 2030.

The price forecasts are subject to high uncertainty, reflected in the large, implied bounds. Large confidence bands (we represent 40% highest posterior density credible sets) may originate from the uncertainty about the reduced-form VAR coefficients, measurement errors in the historical data, uncertainty about other future shocks influencing the price along the forecast horizon (we show the distributions of future shocks in the online-appendix), and the uncertainty around the structural impact effects of the different shocks. In general, confidence bands around structural scenario forecasts are rather large (compare the applications on monetary policy and bank profitability stress-testing in Antolín-Díaz et al., 2021).

Another source of uncertainty, which we do not model directly, is the uncertainty that surrounds the consumption scenarios. First, demand for each metal will depend on technological change that is hard to predict. Second, the speed and direction of the energy transition depend on policy decisions that can have a major impact on metals consumption.

4.3 Revenue Forecasts

We estimate that the energy transition could provide significant windfalls to metals producing firms’ and countries in the net-zero emissions scenario. The potential metal demand boom could lead to a fourfold increase in the value of metals production, totaling US$ 13.0 trillion accumulated over the next two decades for the four energy transition metals alone, providing potentially significant windfalls to commodity producers (see Table 3). Most of it would come from copper and nickel, but the revenues from lithium and cobalt could also be substantial.

^{Table 3:}

Estimated accumulated value of global metal production from 2021–2040. Note: Estimates are in billion 2020 USD.

^{Table 3:}

Estimated accumulated value of global metal production from 2021–2040. Note: Estimates are in billion 2020 USD.

	Historical (1999–2018)	Stated Policy Scenario (2021–2040)	Net-Zero Scenario (2021–2040)
Selected Metals	3,043	4,974	13,007
Copper	2,382	3,456	6,135
Nickel	563	1,225	4,147
Cobalt	80	152	1,556
Lithium	18	18	1,170
Fossil Fuels	70,090		19,101
Crude Oil	41,819	-	12,906
Natural Gas	17,587	-	3,297
Coal	10,684	-	2,898

View Table

^{Table 3:}

Estimated accumulated value of global metal production from 2021–2040. Note: Estimates are in billion 2020 USD.

	Historical (1999–2018)	Stated Policy Scenario (2021–2040)	Net-Zero Scenario (2021–2040)
Selected Metals	3,043	4,974	13,007
Copper	2,382	3,456	6,135
Nickel	563	1,225	4,147
Cobalt	80	152	1,556
Lithium	18	18	1,170
Fossil Fuels	70,090		19,101
Crude Oil	41,819	-	12,906
Natural Gas	17,587	-	3,297
Coal	10,684	-	2,898

The estimated value of production of these four energy transition metals alone would rival the estimated value of crude oil production in the IEA’s net-zero emissions scenario (see Table 3). It would still be substantially below the total value of all fossil fuel production.¹¹ It is also important to keep in mind that there are other metals that would be impacted by the energy transition.

More specifically, Figure 5 shows that the revenue would strongly increase during the 2020s but then either flatten out or even reverse in the 2030s, as supply adjusts for all metals except lithium. Annual copper revenues would more than double from around US Dollar 150 billion in 2020 to more than US Dollar 350 billion in 2030. The nickel market

View Full Size

Annual revenues (in real US$) in the stated policy and the net-zero emissions scenario with 40% point-wise credible sets.

Citation: IMF Working Papers 2021, 243; 10.5089/9781513599373.001.A001

• Download Figure
• Download figure as PowerPoint slide

Cobalt and lithium markets are currently comparatively small with annual values of US Dollar 4.9 billion and US Dollar 2.3 billion, respectively. However, the relative increase would be much larger for these two minor but rising energy transition metals. For cobalt, annual revenues reach a peak of US Dollar 129 billion in 2030 in the net-zero emissions scenario. Cobalt production revenues could decline afterwards due to the decreasing scenario price from 2027 on-wards as supply re-adjusts. Annual lithium revenues would steadily increase by a factor of 50 to US Dollar 117 billion in 2040. In the stated policy scenario estimated revenues would increase moderately to historical highs.

5.1 Alternative Anchor Variable

We replace the real cotton price with real prices for barley and coffee, respectively (both sourced from Jacks and Stuermer, 2020), and also a historical US equity total return series from Jordá et al. (2019). Results are robust, showing no large deviations for neither the estimated elasticities, the maximum prices nor for the estimated revenues compared to our baseline.

^{Table 4:}

Sensitivity: Copper model. Note: US Dollar (USD) refers to real 2020 prices, adjusted for inflation based on the US-CPI.

^{Table 4:}

Sensitivity: Copper model. Note: US Dollar (USD) refers to real 2020 prices, adjusted for inflation based on the US-CPI.

	Elasticities		Scenario Analysis
	Impact	20 Years	Max Price USD per mt	Total Revenue Tril. USD
Baseline	0.35	0.75	9,861	6.1
Altern. 4th variable
Barley	0.33	0.79	8,534	5.2
Coffee	0.33	0.75	8,760	5.5
Equity Index	0.27	0.66	9,545	5.7
Altern. Econ. Act. Var.
Global Real GDP	0.22	0.16	21,570	11.9
Altern. Sample & Trend
1880–2020, no trend	0.40	0.81	8,164	5.2
1955–2020, with trend	0.18	0.27	20,520	11.7
1955–2020, no trend	0.19	0.28	18,200	9.6
3 Variables Model	0.35	0.85	8,301	5.1

View Table

^{Table 4:}

Sensitivity: Copper model. Note: US Dollar (USD) refers to real 2020 prices, adjusted for inflation based on the US-CPI.

	Elasticities		Scenario Analysis
	Impact	20 Years	Max Price USD per mt	Total Revenue Tril. USD
Baseline	0.35	0.75	9,861	6.1
Altern. 4th variable
Barley	0.33	0.79	8,534	5.2
Coffee	0.33	0.75	8,760	5.5
Equity Index	0.27	0.66	9,545	5.7
Altern. Econ. Act. Var.
Global Real GDP	0.22	0.16	21,570	11.9
Altern. Sample & Trend
1880–2020, no trend	0.40	0.81	8,164	5.2
1955–2020, with trend	0.18	0.27	20,520	11.7
1955–2020, no trend	0.19	0.28	18,200	9.6
3 Variables Model	0.35	0.85	8,301	5.1

^{Table 5:}

Sensitivity: Lithium model. Note: US Dollar (USD) refers to real 2020 prices, adjusted for inflation based on the US-CPI.

^{Table 5:}

Sensitivity: Lithium model. Note: US Dollar (USD) refers to real 2020 prices, adjusted for inflation based on the US-CPI.

	Elasticities		Scenario Analysis
	Impact	20 Years	Max Price USD per mt	Total Revenue Tril. USD
Baseline	1.69	2.55	15,724	1.2
Altern. 4th variable
Barley	1.54	2.30	14,475	1.0
Coffee	1.57	2.53	14,930	1.0
Equity Index	1.46	2.32	15,629	1.1
Altern. Econ. Act. Var.
Global Real GDP	1.75	2.82	13,119	0.9
Altern. Sample & Trend
1955–2020, trend	1.69	2.59	19,227	1.5
3 Variables Model	1.57	2.41	15,030	1.1

View Table

^{Table 5:}

Sensitivity: Lithium model. Note: US Dollar (USD) refers to real 2020 prices, adjusted for inflation based on the US-CPI.

	Elasticities		Scenario Analysis
	Impact	20 Years	Max Price USD per mt	Total Revenue Tril. USD
Baseline	1.69	2.55	15,724	1.2
Altern. 4th variable
Barley	1.54	2.30	14,475	1.0
Coffee	1.57	2.53	14,930	1.0
Equity Index	1.46	2.32	15,629	1.1
Altern. Econ. Act. Var.
Global Real GDP	1.75	2.82	13,119	0.9
Altern. Sample & Trend
1955–2020, trend	1.69	2.59	19,227	1.5
3 Variables Model	1.57	2.41	15,030	1.1

5.2 Alternative Economic Activity Index

We use annual data on global real GDP instead of the freight rate index as a proxy for global economic activity. We plot the two series in the online-appendix. The disadvantage of using GDP over freight rates is that it also includes the service sector and that reliable historical data is only available for a small subset of countries. Both factors may bias our results. Unlike the freight rate index, it is also not a leading indicator of metals demand. On the plus side, it seems to better represent movements in economic activity during the Great Depression period than the freight rate index.

The results based on a model with global real GDP show lower estimated elasticities. In particular, the estimated long run elasticity is lower than the front year one in the case of copper. As a result, maximum prices and revenues for copper, nickel and cobalt are substantially higher for the median compared to the baseline. The results are about the same for lithium. The online-appendix shows the impulse responses from the different models using global real GDP.

We chose the results based on the freight rate index as our baseline due to its more favorable characteristics, but also because results are more conservative in terms of price and revenues scenarios. However, we note that the risk is to the upside based on the results for this alternative variable for economic activity.

5.3 Alternative Trend Specification

We chose to include linear trends in the copper and nickel regressions due to their relatively long sample periods. In contrast, we did not include linear trends into the specifications for cobalt and lithium with its shorter sample periods.

The estimated supply elasticities are quite robust to the inclusion or non-inclusions of these linear trends across all four metals. The estimated maximum prices and revenues are also quite robust in the case of copper and lithium but show some sensitivity for nickel and cobalt.

There are negative trends in output for both nickel and cobalt. While yearly production growth averaged 7.1% for nickel since 1900 and and 6.5% for cobalt since 1925, yearly average growth rates decreased to 3.5% and 4.9% over the last 30 years, respectively. That explains why the estimated maximum price and revenues are lower when not including linear trends. The models yield unconditional forecasts with higher production growth rates in this case. In contrast, including a linear trend leads to lower production growth in the unconditional forecast and therefore to a higher estimated maximum price and revenues. Due to the shorter sample for cobalt and the smaller change in average growth rates over the years, we report the baseline cobalt model without a trend.

5.4 Alternative Sample Period

Using a long sample period allows us to cover multiple periods of booms and busts in the metals markets and to get a more solid foundation for the scenario exercise. However, we still check for the robustness of results based on a shorter period, starting in 1955, the same year that the available data for lithium starts.

Sensitivity results show that based on the shorter sample period, elasticities are substantially lower, and prices and revenues are higher compared to the longer sample periods. One reason for this is that growth rates of output tend to be smaller in the later parts of the sample. In the case of nickel, an upward trend in prices, driven by the last decade may play an additional role. For cobalt the short sample seems to make results very sensitive to the inclusion of a trend. As the sample starts in 1955, it only includes 65 observations, covering only a few periods of boom and bust in prices. Plus, lower degrees of freedom make these estimates less reliable. Longer sample results are preferable for our twenty-year scenario horizon. However, we note that the price risk is to the upside based on this set of sensitivity results.

5.5 The Three-Variables Model

Finally, we compare our baseline four-variables model to the standard three-variables commodity-market model without the anchor-variable (e.g., Kilian, 2009, Kilian and Murphy, 2012, Baumeister and Peersman, 2013, Jacks and Stuermer, 2020), using the same narrative sign restrictions (as in Table 2). Results are very robust for the estimated supply elasticities based on the metals-specific demand shock as well as maximum prices and total revenues.

Table 3 in the online-appendix displays the sign restrictions used to identify an aggregate metal demand shock, a metal supply shock and a metal-specific demand shock. The disadvantage of the three-variables model is that we need to assume that there is a negative impact on global economic activity within the first year, which is not fully grounded in economic theory of the energy transition. On the one hand, the energy transition might be interpreted as a negative supply shock (cost-shock) that makes parts of the capital stock obsolete and sees workers reallocate to renewable energy sectors. On the other hand, technological change makes renewable energies significantly cheaper (Acemoglu et al., 2012) and in the long-term global economic activity may benefit. The identification restriction of a negative effect of the metals-specific demand shock on economic activity is, however, necessary to differentiate between the aggregate and the metal-specific demand shock in the three variable VAR.

The impulse responses are shown in the online-appendix. The effect of the metal-specific demand shock on economic activity is slightly stronger in the three variables model and more persistent. Here, the shock significantly lowers economic activity, while the shock is less persistent and in most cases only borderline statistically significant (for nickel, cobalt and lithium) in our baseline model.

6 Conclusion

We examine to what extent metals critical to the energy transition may become a bottleneck. We estimate that the elasticity of supply of key energy transition metals is low in the short run but higher in the long term, especially for lithium. Based on metal-specific demand shocks embedded in a structural scenario analysis, we find that prices of lithium, cobalt and nickel could rise several hundred percent compared to their average 2020 levels in a net-zero emissions scenario, representing a major bottleneck. The four metals prices would roughly increase to historical peaks in real terms, but remain there for a longer period of time than seen before. Robustness checks suggest that there is upside risk to net-zero emissions scenario price paths. Metals producers of these four metals alone could generate revenues similar to those of the oil sector over the next 20 years in this scenario.

In addition to contributing to the literature on metals supply elasticities, our analysis offers a novel identification approach of commodity-specific demand shocks using an “anchor” variable that increases resemblance of this shock to an energy transition induced metals’ demand increase.

Our model assumes that supply elasticities stay constant in the future. These elasticities could be higher due to technological change or economies of scale, as firms figure out faster ways to expand supplies through mining but also through enhanced recycling. At the same time, the environmental and social costs of mining could also decrease these elasticities in the future. Our robustness checks suggest that elasticities are lower for most metals in the more recent part of the sample. Overall, we believe that a constant elasticity is a balanced assumption for the scenarios.

There are a number of potential policy implications of our results. First, if the energy transition is surrounded by high uncertainty, this could delay investment in metals production and supply may not adjust in time. A credible, globally coordinated climate policy that directs investment to sufficiently expand metals supply could, hence, play a decisive role in avoiding unnecessary cost increases of low-carbon technologies. Countries may want to announce slow but rising commitments.

Second, a substantial expansion of mining could have adverse effects on the environment, unless firms adopt mitigating technologies. Stringent high environmental, social and governance standards are therefore important. Incentivizing recycling, reuse, and refurbishment as well as metal efficient product designs is also vital part of the energy transition.

Third, the energy transition will create winners and losers, potentially requiring fiscal or structural policy interventions. Commodities exporting and importing countries may be affected differently by the energy transition, depending on the scenarios and the behavior of specific metal markets and prices. Specific guidance may be needed for countries benefiting from large windfalls (e.g., establishment or strengthening of reserve funds, frameworks on how to share the gains).

Fourth, some metals have already been subject to export restrictions as well as subsidies for domestic mining. An accelerated energy transition may lead to additional trade restrictions, which could cause additional cost increases and hamper investments in clean energy technologies. Reduced trade barriers and more stringent multilateral rules on export restrictions would allow markets to operate efficiently.

Finally, the creation of an international institution focused on metals—analog to the International Energy Agency for energy and the Food and Agricultural Organization for agricultural commodities—could play a pivotal role for data dissemination and analysis, setting industry standards, and be a forum for international cooperation.