This is the final article of a 6 part series on the subject of exploration risk. Each part references subjects from previous posts and so should be read sequentially. The covered topics are:


In the previous posts I detailed the stages needed to culminate in a statistical model of exploration risk that can be used to make investment decisions and guide exploration management. In this post I hope to look in a little more detail at how that model works and continue some of the criticism of the model from the previous post. The results will be used to infer some initial conclusions about the current market and the way the exploration industry works.

Impact of ‘World Class’ to Small Deposits

When looking at the model it is apparent that the model rarely creates resources in excess of 10Moz, reflecting their scarcity in nature. However the model does have the capacity to create rare huge resources, though I capped at 65Moz (conceptually similar to a resource estimation top-cut). From 100,000 simulations, 14 resources over 20Moz were created which is estimated to account for ~10% of the total simulated value. The majority of value (>65%) came from <4Moz mines. The single largest resource simulated at 65Moz accounted for 4% of the total simulated value. Current annual production of gold is ~90Moz pa and gold from mines with resources >20Moz is ~11%, the largest producer (Muruntau) accounting for ~3%. It would therefore seem this model produces representative simulations. Natural Resource Holdings produced a nice infographic breaking down some of the current distribution of global gold (Natural Resource Holdings. 2014), which beaks down some of the statistics of global gold production and I recommend readers take a look. The model fits many of the observations in the Natural Resource Holdings infographic well despite the fact that it is restricted to a single gold belt. This indicates that trends seen in the Abitibi may well be replicated in other mesothermal gold belts. It’s similarity with global gold distribution indicates that natural distributions in mesothermal belts may be similar in nature so distributions in other gold regions such as epithermal belts, however I certainly feel caution should be exercised in using such extrapolations over deposit types that differ so radically geologically, geographically and temporally.

Modelled Distribution of Resources

News Post 3

That resources >20Moz account for a significant portion of the value is an important realisation. It would be very tempting to lower the maximum resource size to something more commonplace, perhaps 20Moz, seeing as only 0.01% of simulated resources are above this threshold. However this would disregard one of the main reasons why the participants in exploration put up with the massive risks and that is the narrow tail end chance of discovering a mega-resource.

In the model I included a minimum resource size of 0.75Moz for low grade resources and 0.3 Moz for high grade resources. The minimum resources reflect the fact that for hard rock operations there is a minimum resource size below which economics get squeezed and mines unfinancable. 3.8% of simulations resulted in a resource that was large enough to be successfully drilled and produce a resource but was either below these thresholds or uneconomic. This compares with 0.65% of simulations that produce an economically viable mine. Thus we can see that there is lots of potential for non-commercial occurrences of gold to create gold targets that are worthy of drilling but are ultimately uncommercial. One of the challenges of the exploration is to try and assess the commerciality of an undrilled prospect but with such limited knowledge this will always be a process fraught with uncertainty, not least because many of these deposits are on the margins of commerciality.


Discovery volatility was set by the development level of the project and the industry standard accuracy levels (equivalent to SD or Dσ) associated with that level. The volatility was modelled backwards from the value of R in year 6 and flexed 10% in feasibility in year 5, 30% in PFS in year 4, 50% in Scoping Study in year 3 and 75% in the Maiden Resource in year 2. On successfully drilled projects, Dσ over the project cycle is therefore ~95%. This reflects the risk of under or overestimating (to the possible extent of not even identifying an existing viable deposit) what’s there from initial drilling. Intimately related to Dσ is the chance of success in the first 2 phases of work, 10% that I’ve ascribed to chance of success in reconnaissance (phase 1) and 45% to drilling (phase 2). I’m using reconnaissance here as a byword for exploration processes that produce & prioritise drill targets, often geochemistry & trenching but in areas of deep cover possibly geophysics or RAB drilling. A hit rate of 1 in 10 for this reconnaissance phase is based on my experience in the industry as a whole. Arriving at more statistically justified figures for this hit rates is something that needs proper statistical scrutiny. As discussed in the previous post the 45% hit rate of drilling success is measured from study of drill results of companies operating in the Abitibi belt.

Notwithstanding the estimated nature of the reconnaissance hit rate it is clear that the model is estimating a great deal of variability through the exploration process post phase 2 from out application of Dσ. Again this is an area that could benefit from greater research. Particularly informative would be research into exactly how accurate are the various levels of technical study during exploration. If we are to believe the assumptions as stated above then it is clear that significant risk still remains after initial discovery. Market euphoria can cause severe spikes in value in the aftermath of a discovery and savvy investors may wish to use this as a chance to take profit as much risk still remains to be resolved.

The model utilises a single Rσ value which is resolved at t=0 and does not change. This is true to nature, after all in any mineral deposit new mineralisation isn’t created it is just discovered and this process was accommodated by Dσ. In the model Dσ is reduced to zero at the end of discovery phase and we enter mining. Though this is a neat way of breaking up our different development phases it is not strictly true as rarely does a mine reconcile precisely to its resources and reserve estimates at the time of construction. In reality the discovery process is present to some degree until the mine finally shuts. This is yet another area that requires further study but was outside the scope of these posts. It would be very interesting to see the distribution of mined gold in relation to estimates at opening to see if it adds an additional degree of optionality value to mining projects.

Gold Price:

Current price is taken from the spot price and volatility from previous annualised price movements. I also added a price ‘floor’ of $800/oz, other practioners may choose not to put in such a floor. Conversely a price cap may also be considered by some. Using the inputted variables gold price varied from $800 to $3676, with 98% of simulations resulting in a value of less than $1920. Commodity price is always a difficult variable when calculating mining project value but is commonly taken as a static figure. This approach is inherently conservative (unless a price way over spot is used!) as it ignores the fundamental optionality of such projects. Mines normally have an option to delay or shut down operations in some way and this limits the downside risk such a project has but leaves the upside uncapped. Such option like behaviour can be significant and should be considered for a full picture of project value. The 2% of projects simulations with a gold price above $1920 accounted for 8% of the total simulated value which demonstrates the optionality value of exposure to commodity price inherent in exploration.

$/oz metrics:

I made the decision in the model to value the project at the end of year 2 through to feasibility via $/oz metrics which I think reflects the fact that such projects are often sold to a larger mining company and such deals often utilise such metrics. The metrics were flexed in accordance to grade of the deposit, level of development and gold price such that the metrics increased with increasing gold price. For simplicity projects were considered as either high or low grade, below is the table of metrics used at a gold price of $1100:

$ per ounce in the ground metrics at $1100 gold:

Project Level Low Grade High Grade
Feasibility Level $50 $90
PFS $40 $75
Scoping Study $25 $50
Maiden Resource $10 $20

In practice these values will change on a number of factors, not least depending on level of certainty in resource (inferred/indicated/measured), however for the purpose of quickly identifying potential values in a model with thousands of iterations the metrics were simply multiplied against the expected value of R at that stage.

Exploration & Mining costs:

These were broken down into a minimum value for each year and a floating component. The minimum value reflects that even a technical study on a tiny resource needs a minimum number of technically competent people and therefore has a minimum cost. The floating component adds up to a somewhat industry standard of $20/oz (based on other authors work on global exploration spend vs discovered ounces) which is apportioned to the various stages past the maiden resource with weighting towards the feasibility in year 5. In 2012 the average global cost to discover an ounce of gold was ~$30 (close to the average for Canada), I have deducted $10 from this total due to the model suggesting that for every ounce discovered, $10 is spent unsuccessfully exploring other prospects that never get to the resource stage.

There has much been made in the past decade of the increasing cost of finding new deposits and our seemingly declining exploration performance as an industry generally. This theory would appear sound in that as time passes, good deposits, at shallow depths, in safe jurisdiction and legally unencumbered, are scarcer and difficult to find. A more refined model may try to encapsulate this exploration cost escalation and the effects of declining resource abundance, particularly on a terrane level, this is dealt with in greater detail in the discussion section.

In year 6 of the model I create a simplistic NPV model that requires estimation of mining costs. The model assumes that all the capex is consumed at year 0 and that production begins after that year without ramp up (though a ramp up period could be incorporated relatively easily). Gold price was not flexed as without flexing mining costs (a possible future refinement) it may introduce bias as the 2 are correlated i.e opex per ounce can change in different price environments as COG and with it head grade changes. I estimated capex into a fixed and floating component to reflect increased expenditure on larger mines. I used the grade and resource size to infer a tonnage (TPA), mine life (max 25 years) and throughput. Operating costs are applied per tonne of ore and varies depending on if the TPA is small, medium, or large. I also applied a fixed operating cost to reflect the corporate overhead required for even small operations. I applied a global recovery rate of 85%, inclusive of both mining and mill recovery.

Discussion and Conclusions:

A major criticism of the model created is that it assumes discovery rates remain constant through time, which we know is inaccurate. In the Abitibi many of the most major discoveries were made in the earliest stages of the provinces exploration, and it therefore may seem odd to use past discoveries to guide our expectations of future ones. This is a valid criticism but does not mean this model is without value. It is hoped that there is enough similarity between terranes of similar nature that models created from study of mature districts can be used to guide expectations of discovery on virgin districts. In this respect readers may be interested in the thinking espoused by Richard Sillitoe (2012) amongst others. Though thinking about copper (I believe the same can be applied to gold), Sillitoe postulates that certain parts of the globe have a greater endowment for reasons not yet fully understood put possibly from subduction and other tectonic events in the ancient geological past and predating the main mineralising epochs. If we are to apply such theories of mineral inheritance it becomes harder to justify the use of the endowment of one region to make predictions about another, however for lack of usable information, models such as we've created though flawed may be the best guidance we have.

As stated in the previous post the average value of a prospect at time 0 is $1.1M. This is a lot higher than mineral companies are currently valued at in the market given that the average company would usually have multiple prospects. This and the relatively high success ratio of 0.65% compared to industry rules of thumb of 0.1% to 0.01%, may in some way be explained by the previous paragraph. Much exploration is conducted in well understood terranes where we may expect that the majority of large deposits have already been found, which in turn must effect our assumptions of Rσ. We can see from our analysis that those mines >4Moz contribute to over a third of the total value of successful projects and so any reduction of this ‘long tail’ would be significant. If in mature districts such as Abitibi these discoveries can no longer be made then over a third of the provinces value has gone, not even accounting for the value lost in the deposits <4Moz that have already been discovered.

In effect the Rσ would be expected to reduce in more mature terranes reducing the value of resource volatility. As we saw in post 2 the effect of reducing such volatility is a reduction of value of the project as a whole. As the amount of expenditure spent on unsuccessful exploration would only increase the resulting effect is that for mature terranes the $1.1Moz value must be lower, in part explaining the comparatively low market valuations. This also gives some insight into the magnitudes of what is termed ‘first mover advantage’ or exploring in previously unexplored areas. Such exploration does not regularly achieve higher market valuations, often because investors are rightly worried if the unexplored district can be host to a new mineral province. Models such as we have constructed can be used to guide what level of risk may be accepted to invest in such new province propositions, which if proven to exist would be expected to have higher values of Rσ and a correspondingly higher valuation. For mature terranes the same models can help us think in more depth about the effects of resource depletion on a district and the sorts of exploration risk we may be willing to accommodate given the reduced value of Rσ.

There is much more work needed to be done to ascertain how accurately the market values exploration and commodity risk but it is hoped exercises such as those considered in the previous 6 posts can begin to give us insights into the true nature of these risks. Once there nature is understood more accurate valuations for exploration property can be generated and compared to the markets valuation, hopefully improving the industry’s capital deployment into exploration as a whole. Traditional valuation methods used in later development/mining stage projects tend to be more conservative in nature than the methods we have looked at, and for many investment decisions being conservative is a positive. Methods like we have looked at do not necessarily need to completely replace this conservative approach, however to get a more complete picture I think there is room for both types of techniques to be deployed in all stages of exploration and mine development.

I hope you have all enjoyed reading my posts on this subject, I hope to write more in the future on some of the issues discussed.


Sillitoe, R.H. 2012, Copper Provinces. SEG Special Publication No.16, pp 1-18. SEG.

Desjardins, J. "Global Gold Mine and Deposit Rankings 2013 - Visual Capitalist." Visual Capitalist. Natural Resource Holdings, 8 Feb. 2014. Web. 18 Nov. 2015.