Estimation methodology

The first step is to estimate the amount of metal produced in metric tons (tonnes). The transformation from US dollar revenue to tonnage is carried out using the spot price of each respective metal. The Bloomberg tickers used as a proxy for the spot price are given in Figure 4. Since the price varies over the course of the year, we use the average spot price per calendar year. In the absence of a more detailed picture of hedging behavior, this assumption is reasonable.

Figure 4: Bloomberg tickers: Metal spot prices

Source: Bloomberg

Based on the spot prices, we have a corporate dataset containing an estimate for physical production/extraction by metal and the overall GHG emissions. The output varies considerably by firm as does the product mix (Figure 5 and 6). For this analysis, we do not account for possible economies of scale in production (which is typically modeled using a version of the Cobb-Douglas production function) since we want to keep the regression model parsimonious. We might revisit this assumption in later publications.

Companies are divided into production groups—which are defined by the mix of metals each company produces. As seen in Figure 5, we have 23 production groups. A pre-processing step for the regression is to calculate an average emission per metal based on each of the production groups.

Figure 5: Pure-plays versus mixed production

Source: Bloomberg

Our research has determined that limiting cross-production produces more credible estimates. Accordingly, we start by focusing on the industrial metals sector (hence excluding companies that produce any precious metals).

Figure 6: Estimated metals & GHG production (Base-only groups, 2012 - 2019)

Source: Bloomberg

Calculating metal intensities

Given the use of annual data and the steady increase of GHG disclosure, we use the full sample for the analysis. In our study, the estimation of metal intensities is a based on a three-step process:

1. Gold intensity is based on pure-players, using a trimmed mean;
2. Industrial metals intensities are calculated using OLS regression on data aggregated to production groups; and
3. A plug-in approach is used to calculate the silver intensity using the gold-silver joint producer group.

Gold intensity

Since we have a relatively large number of pure-play gold miners, we estimate intensity by using the median of intensities across companies. This accounts for outliers with negligibly small production levels. The (median) gold intensity is calculated to be 20,298 tons of GHG per (metric) tonne of gold extracted.

Industrial metal intensity

The regression estimates for industrial metals are given in Figure 7 along with the confidence intervals. As we see, the estimation is noisy around the mean — leading to intervals that contain negative values in the case of copper, nickel and zinc.

Figure 7: Base metal intensifies

Source: Bloomberg

Naturally, the floor on estimates are 0 and we show the negative values simply to highlight the uncertainty bounds around the estimation. We believe these are a reflection of the variation due to disparate geographies in which the metals are mined, differences in infrastructure and the difficulty in attributing emissions to individual metals in cases of co-production.

Silver intensity

Due to the fact silver is often extracted as a by-product of zinc, nickel and copper mining, it is difficult to estimate the intensity using the standard regression-based approach. Instead, silver intensity is estimated as a residual using the emissions data from the joint gold-and-silver producer group. The gold estimate (from above) is plugged-in to provide the gold contribution to GHG emissions. The residual amount of emissions is attributed to silver extraction. Based on this method, the median value of the silver intensity is 63.9 tons of GHG per metric ton of silver.

How do the estimates compare to the literature

By highlighting the emissions over the lifecycle of the extraction process, the body of academic literature illustrates the different considerations that lead to estimation variability. These include (in no particular order) the location of mines (soil/rock composition), method of extraction and refining, the equipment used for extraction, energy sources and the percentage of recycling. This is reflected in the confidence intervals (Figure 8).

Figure 8: Literature estimates

Source: Bloomberg, The Silver Institute, The World Gold Council, see references for other publications

Some common examples include the dominance of renewable energy versus coal in Scandinavia and Australia respectively, the use of different purification processes for refining zinc and the lower emissions levels of using recycled scrap metals versus primary mining.

Aggregating over different geographies and companies creates an average value that might not fit many companies individually, but is the best representation of the group. This should be consistent with a commodities investor’s needs—in that the reference (deliverable) entity for a commodities futures contract is not linked to a particular company; and hence can treated as a (hypothetical) ‘average producer’.