ISSUE

43
November 2024
The Alchemist
Outsmarted by gold…
Introduction Consensus Forecasts

Imagine commodity price forecasters as Dr Strange in The Avengers In this analysis, we examine the accuracy of gold price forecasts as
Endgame, sifting through an infinite number of possible futures. Like compiled by Consensus Economics, a widely recognised aggregator of
the ex-surgeon turned mystic hero, they must juggle all potential commodity price forecasts. Consensus Economics gathers predictions
outcomes, trying to piece together a recommendation that won’t from over 30 leading global forecasters each month, and by averaging
implode. In the wild world of mining and commodities, their these forecasts, a single consensus forecast is produced, designed to
superpower? Navigating the chaos of commodity price predictions. incorporate multiple market perspectives and forecasting
While the Avengers manage to nail their 1 in 14 million shot at victory, methodologies.
the real question is: How does the industry fare when it comes to
cracking the code on commodity price forecasts? We’d argue that it’s To assess the performance of these forecasts, we selected the period
at longer odds than The Avengers…. from January 2018 to October 2024 for it is a mix of market conditions,
including relatively stable range bound periods, and bull runs. While the
Accurate commodity price forecasting is essential for both public and data includes instances of price declines exceeding 10% (notably
private sectors. These forecasts are used in resource estimates and between August 2020 and March 2021), no structural bear markets are
mining feasibility studies, influence government policies and fiscal considered present in this period, although it could be argued that gold
budgets, and impact entire economies by affecting potential future has not seen this for a decade where in the period between 2012 and
trade balances. 2015 gold fell by approximately 40%.

In this issue of the Alchemist, we take a deep dive into gold price This analysis is not intended to evaluate the accuracy of individual
forecasts and utilise a number of statistical tools, analysing their forecasting groups or their methods. Instead, it aims to examine how
performance. effectively the industry forecasts gold prices overall, given its critical
nature.

Key Findings of The Alchemist Issue 43

How has the forecasting industry performed?

1. Consensus forecasts are susceptible to anchoring bias. To visualise gold price forecasts alongside actual performance, the
Forecasts often “anchor” to the current spot price, closely data was aggregated by monthly forecast and plotted in Figure 1. The
following short-term price movements rather than accurately actual gold price is shown in red, while forecasted prices are
predicting significant shifts or volatility. This leads to a represented by dotted lines.
persistent bias that has continually underestimated both
upward trends and the true volatility of the gold market. Over the past five years, it is apparent that consensus forecasts have
consistently underestimated the actual gold price. Forecasts appear to
2. Consensus forecasts lack statistical accuracy in have projected a continuation of prevailing prices, struggling to capture
predicting gold price trends. Statistical analyses reveal the impact of volatility or significant upward movements. While there is
that consensus forecasts frequently fall short as predictors of a common perception that forecasts merely “follow the spot price,” in
actual gold price movements. They often project a flat or practice, it appears that during volatile periods, forecasts simply adjust
declining trend, which contrasts with the observed price in response to spot market shifts. As they move with the spot price, they
increases over time. tend to predict a reversion to historical base levels. As the spot price
stabilises, forecasts seem to establish new baselines, suggesting that
3. Forecasts typically predict a declining nominal gold gold price forecasts have not effectively accounted for volatility over the
price while actual historical gold returns are positively past seven years.
weighted. Historical data since the 1980s shows an average
annual change of +4.9%, with a standard deviation of 15.5%, A notable observation from Figure 1 is the high number of forecasts
reflecting a long-term upward trend. This positive weighting falling below the spot price at the equivalent time, highlighting the
contrasts sharply with the conservative, often flat or declining magnitude and persistence of the issue. As an aggregate 80% of all
forecast profiles, suggesting that forecasting models fail to predictions when assessing all periods underpredict the gold price. This
fully incorporate gold’s historical upward momentum, is clearly illustrated in the huge, persistent area of dotted forecast lines
particularly over longer time frames. UNDER the red spot price line compared to those above. Further, at
the 12-month mark, 81% of forecasts have underestimated the gold
price, increasing to 90% at the 20-month mark.

1
RFC Ambrian Limited
 ISSUE

43
November 2024
The Alchemist
2,800

2,600

2,400

2,200

2,000
US$/oz

1,800

1,600

1,400

1,200

1,000

Gold Spot Price

Figure 1. Consensus gold price forecasts (shown by dashed lines) since 2018 vs spot price

2
 ISSUE

November 2024
43 The Alchemist
Black Panther and the Anchoring Bias?

The concept of “Anchoring Bias” originated from Amos Tversky and Plotting these outcomes on a smoothed histogram indicates that
Daniel Kahneman's 1974 paper titled "Judgment under Uncertainty: consensus gold price forecasts generally vary within a ±10% range of
Heuristics and Biases". In their seminal work, they noted that people the initial spot price across most forecast intervals (see Figure 2). A
often make estimates by starting from an initial value and then adjust it shift from a neutral to a slightly negative bias between 6 and 12 month
based on a limited set of heuristic principles. This approach simplifies forecasts can be observed, suggesting that, over longer horizons,
the estimation process but can lead to inaccuracies, often because the forecasts tend to predict negative nominal price movements for gold.
predictor or model makes insufficient adjustments from the initial
“anchor” point, resulting in inaccurate forecasts and, specifically, an In contrast, actual gold price movements, shown in Figure 3, display a
underestimation of volatility. distribution with wide tails that expand over longer intervals.
Additionally, actual performance is positively skewed, and this skew
To assess the presence of anchoring bias, an analysis of consensus intensifies over extended time frames.
forecast gold price performance versus actual gold price performance
can be conducted. If anchoring bias is present, one would expect an This analysis highlights a form of Anchoring Bias in consensus
underestimation of volatility, with forecast price distributions clustering forecasts, an overreliance on a “safe” view and starting point that limits
around the initial gold spot price. both adaptability and volatility. Much like Marvel's Black Panther, who
initially restricts Wakanda’s potential by clinging to the “safe” and
To measure forecast accuracy, the following equations were used: stable, but outdated, tradition of isolation, consensus price forecasts
appear to remain tethered to the “safe” initial spot price. This
𝐹𝐹𝐹𝐹𝑛𝑛 =
𝐹𝐹𝑛𝑛 − 𝐺𝐺0
𝑎𝑎𝑎𝑎𝑎𝑎 𝐺𝐺𝐺𝐺𝑛𝑛 =
𝐺𝐺𝑛𝑛 − 𝐺𝐺0 conservative bias in forecasting leads to an underestimation of price
𝐺𝐺0 𝐺𝐺0 volatility and an, ultimately inaccurate, negatively weighted forecast
distribution, in contrast to the far more volatile and positively weighted
where:
actual price performance of gold.
G0 = Gold price at commencement of forecast period

Fn = Forecast gold price at period n

Gn = Actual gold price at period n

FCn = Forecast gold price performance from spot price at prediction at period n

GCn = Actual gold price performance from spot price at prediction at period n

25
Frequency (months)

20 Month 3

15 Month 6
Month 12
10
Month 15
5 Month 18
Month 20
0
-20% -15% -10% -5% 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65%
Forecast variance between spot gold price at forecast commencement and forecast price at month [n]

Figure 2 Predicted gold price changes based on the consensus forecast, measured at various monthly intervals from the start of the forecast period

RFC Ambrian Limited 3

 ISSUE

November 2024
43 The Alchemist
25
Frequency (months)

20 Month 3

15 Month 6
Month 12
10
Month 15
5 Month 18
Month 20
0
-20% -15% -10% -5% 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55% 60% 65%
Increase/decrease (%) between spot gold price at forecast commencement and actual gold price at month [n]

Figure 3. Actual gold price changes, measured at various monthly intervals from the start of the forecast period

Analysis of Residuals in Forecasting

One way to assess forecast accuracy is through residual analysis. In a separated by k periods, revealing any potential patterns that suggest
forecasting model, residuals represent the difference between missing information in the model (or an “inaccurate model”).
observed values and the model's predicted values. For this analysis,
the residuals of all 80 consensus forecast periods were calculated The LB test, which provides a QLB value, was then used to quantify the
using the following equation: degree of autocorrelation. A large QLB value suggests that
autocorrelation does not result from white noise, as QLB. is essentially
𝑒𝑒𝑡𝑡 = 𝐺𝐺𝑛𝑛 − 𝐹𝐹𝑛𝑛 the sum of all the autocorrelations at various lags, with large values
representing a persistent pattern in the forecast, which is therefore
where: unlikely to be random.
et = Residual
The formulas for autocorrelation and the Ljung-Box test are:
Fn = Forecast price in period n
Autocorrelation at various lags:
Gn = Actual gold price at period n
𝑐𝑐𝑜𝑜�𝑣𝑣(𝑒𝑒̅𝑡𝑡 , 𝑒𝑒̅𝑡𝑡−𝑘𝑘 )
𝑝𝑝̂𝑘𝑘 =
𝑣𝑣𝑎𝑎�𝑟𝑟(𝑒𝑒̅𝑡𝑡 )
In a reliable forecasting model, residuals should behave as "white
noise", exhibiting two key properties: Ljung-Box test for randomness:

1. Uncorrelated Residuals. Residuals should not display 𝐿𝐿

𝑝𝑝̂𝑘𝑘 2
patterns over time. If they are correlated, this suggests that 𝑄𝑄𝐿𝐿𝐿𝐿 = 𝑇𝑇(𝑇𝑇 + 2) �
(𝑇𝑇 − 𝑘𝑘)
the model is missing information that could improve forecast 𝑘𝑘=1

accuracy
where:

2. Mean of Zero. If the residuals' mean deviates from zero, this k = the associated lag
indicates a potential inherent bias in the model (ie
consistently over- or under- estimating values) p̂k = the autocorrelation at lag k

To test these properties, an aggregate approach was used. The QLB = result of the Ljung-Box test
residuals for each forecast period were averaged to calculate the mean
residual for that period. T = sample size

L = the number of lags being tested

The first property (uncorrelated residuals) was tested by measuring
autocorrelation at various lags and then applying the Ljung-Box (LB)
test. Autocorrelation describes the correlation of a series of variables
between two successive time intervals. By adjusting the lag (the
interval), we measure the correlation between values that are

RFC Ambrian Limited 4

 ISSUE

43
November 2024
The Alchemist
To confirm this, the LB test is checked against the expected outcomes To test whether the residuals have a mean of zero, the aggregated
of QLB where the system is assumed to be random (white noise). This residuals were further analysed, as shown in Figure 4. The results
is done by assessing QLB against the Chi-squared (X2) distribution. A confirm that the mean is not zero, with residuals drifting further from
probability value below 0.05 from the X2 distribution would indicate a zero as the forecast period extends, revealing an inherent and
low probability of the calculated QLB value being returned in a truly persistent bias in the model indicating that it tends to systematically
random system, thereby telling us that the actual system being underestimate values. One potential approach to correct this bias
assessed – as represented by the QLB value – is in fact autocorrelated would be to adjust forecasts by adding these mean residuals; however,
and not random. This would reflect a systematic issue in the model. as the gold price evolves over time, repeated re-analysis would be
necessary to maintain accuracy.
Using the calculated residuals, the LB test yielded a QLB of 121.6, a
high value for this test. The probability of obtaining this QLB value is In conclusion, the consensus forecasts do not satisfy the required
essentially zero (2.4x10-21 to be precise). This is far below the 0.05 properties for an accurate forecasting model given the presence of
significance threshold, meaning the residuals can’t statistically be autocorrelation and a non-zero mean, suggesting potential for
considered random “white noise” and the model’s errors likely reflect significant improvements.
systematic issues. To improve the accuracy additional information is
therefore required to be incorporated into the forecast.

1200

1000

800

600
Plus one Std
400
Less one std
Residual Value ($US)

Maximum
200
Average

0 Minimum

-200

-400
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Forecast Period

Figure 4. Average residuals from consensus gold price forecasts at respective forecast period

RFC Ambrian Limited 5

 ISSUE

43
November 2024
The Alchemist
History - Positively Charge Like Thor’s Hammer?

As discussed in the previous sections, consensus forecasts tend to Table 1. Annual performance of gold
track closely with, and are anchored to, the spot gold price. Additionally,
these forecasts have increasingly underestimated gold prices over Parameter Value
longer time horizons. But is this underestimation rooted in historical
gold price trends? Average Return 4.9%

Gaining a comprehensive understanding of historical gold price trends Return Standard Deviation 15.5%
requires looking back beyond just 2018 to see a fuller picture. Much like
the enduring power of Thor’s hammer, Mjolnir, which is challenging to Median Return 3.1%
control but holds immense energy when harnessed, gold’s long-term
returns show an underlying positive trend that forecasts often struggle P80 of Returns 21.1%
to fully capture. By extending our analysis back to the 1980s, we gain
greater insight into gold’s actual performance across a wider range of
P20 of Returns -6.1%
market conditions and this broader perspective could help inform
whether consensus forecasts are justified in their often conservative
assumptions (in both nominal and real terms). Variance of Returns 2.4%

We chose to begin this retrospective in the post-Bretton Woods era of

the 1980s, as by this time the market had adjusted to the final
abandoning of fixed exchange rates and the US$35/oz “Gold Standard”
in the early 1970s, allowing gold prices to reflect broader market forces.

The annual performance of gold, summarised in Table 1 and the

histogram in Figure 5, highlights a key insight: gold’s returns are
positively weighted in the long term as well as the shorter term, much
like the latent power in Mjolnir. Table 1 shows an average annual return
of +4.9% with a standard deviation of 15.5%, further reflected in the
positive weighting of gold price returns in Figure 5. Given this persistent
upward trend, one might expect consensus forecasts to project gradual
price increases. However, as discussed earlier, the opposite holds true
with consensus forecasts often anticipating flat or declining prices, both
in real and nominal terms, underestimating the full potential energy
behind gold’s longer and shorter term trends.

12

10
Positively skewed returns
8
Return Frequency

0
-35% -30% -25% -20% -15% -10% -5% 0% 5% 10% 15% 20% 25% 30% 35%

Figure 5. Distribution of annual return of gold since 1980

RFC Ambrian Limited 6

 ISSUE

43
November 2024
The Alchemist

3,000

2,500
Gold Price (US$/oz)

2,000

1,500

1,000

500

0
1980 1985 1990 1995 2000 2005 2010 2015 2020

Figure 6. Real gold price since 1980

Impact and long-term trends (upward in the case of gold) are essential to
improve the accuracy of gold price predictions and support informed
Commodity prices have far-reaching impacts, from influencing terms of investment and operational decisions in the gold industry.
trade between nations, to guiding decisions in mining projects at all
stages of progress: study, development and operational. Conclusion

Hedging is a particularly high-profile manifestation of this, as not only a In conclusion, while Dr Strange and the Avengers may be better at
common requirement in project financing to reduce lender risk, but predicting the future than the gold price forecasters, the recent surge in
because the real-world impacts of it are obvious for all to see in gold prices has likely drawn more attention to the gold market than to
company reporting of financial performance and position. However, the latest superhero movie.
given the consistent underestimation of gold prices, volatility, and the
positive weighting of actual price performance, reliance on consensus Consensus commodity price forecasts play a critical role in shaping
forecasts for hedging decisions may prejudice project owners and investment, operational, and financing decisions across the gold
increase their project risk (or least limit their upside exposure at best) industry and beyond. However, it’s clear that forecasting gold prices
rather than reducing it. This effect would be especially pronounced in faces significant challenges due to inherent “safety first” biases and
operations that faced underperformance or cost escalations, underestimations of volatility. Anchoring Bias, has in particular, led to
particularly if coincident with rising gold prices where buying contracted forecasts that remain tethered to initial spot prices, failing to adequately
ounces to deliver into a hedge is beyond the company’s financial account for the positive weighting and volatility evident in actual gold
capacity. price performance over time. This shortcoming is potentially especially
impactful given the wide use of consensus forecasts in both public and
Companies seeking to secure financing and to develop operations private sectors, and the influence that it has on decision making.
should consider these factors carefully when optimising their gold
hedging strategies. However, it’s also important to note that Our analysis suggests that the forecasting industry may benefit from
overestimating gold prices within a feasibility study can jeopardise a refining its approach to gold price forecasting. By adopting improved
project’s viability. As highlighted in previous sections, historical methods and critically evaluating past forecasts, forecasters could build
forecasts have often misjudged gold price movements, leaving more resilient strategies for pricing, hedging, and long-term planning,
unutilised information that could improve model accuracy. ultimately enhancing the value of gold investments and operations
across the industry.
Equally important is how the industry uses and interprets commodity
price forecasts. As the old saying goes, “The only person who gets to
bring belief to a meeting is a priest; everyone else must bring data.” It’s
imperative to revisit historical forecasts and establish a feedback loop
to ensure continuous improvement in the industry’s forecasting tools.
Enhanced forecasting methods that better account for volatility ranges

RFC Ambrian Limited 7

 ISSUE

43
November 2024
The Alchemist

RFC Ambrian Limited

[email protected] www.rfcambrian.com

Level 34, Grosvenor Place Tower Level 48, Central Park Octagon Point
225 George Street 152-158 St Georges Terrace 5 Cheapside
Sydney NSW 2000 Perth WA 6000 London EC2V 6AA

T. +61 2 9250 0000 T. +61 8 9480 2500

F. +61 2 9250 0001 F. +61 8 9480 2511

DISCLAIMER

RFC Ambrian Limited (“RFC Ambrian”) has prepared this note to provide general commentary and analysis to professional and sophisticated investors on
resource companies, securities and markets. No part of this report is to be construed as a solicitation, offer or invitation to buy or sell any security and should
not be relied upon in connection with any contract or commitment whatsoever.
RFC Ambrian prepared this report without taking into account the objectives, financial situation or needs of any person. Before making an investment decision
or otherwise acting on the basis of this report you should consult with a professional investment adviser to consider the appropriateness of the advice, having
regard to your objectives, financial situation and/or needs.
This report is based on publicly available information. Although the information contained in this report has been obtained from sources believed to be reliable
and accurate, its accuracy, reliability or completeness has not been verified by RFC Ambrian and is not guaranteed.
Opinions, conclusions, assumptions, estimates, reflections, forward looking statements and forecasts referred to in this report are those of RFC Ambrian alone
and not those of the companies referred to in this report and these companies do not endorse this report. Opinions expressed reflect RFC Ambrian’s judgement
at the date of this report and may change without notice. Forecasts of commodity prices, interest rates, exchange rates and economic growth are subject to
significant change. No representation or assurance is given that any prediction, projection or forecast contained in this report will be achieved.
RFC Ambrian and its related bodies corporate or any of their associates, officers or employees may have interests in securities referred to in this report and
may hold directorships in or provide corporate finance or other services to the companies referred to in this report. Further, they may buy or sell securities of
the companies referred to in this report as principal or agent, and as such may effect transactions which are not consistent with any opinions contained in this
report.
Use of the information in this report is at your own risk. RFC Ambrian is not responsible for any adverse consequences arising out of the use of this report. To
the extent permitted by law, RFC Ambrian accepts no responsibility for damages or loss relating in any way to any errors or omissions in any information or
opinions provided in this report, whether arising from negligence or otherwise from the use of or reliance on this report.
In Australia this report is intended only for publication and distribution to professional and sophisticated investors and is not to be read or relied upon by any
other person.
This report is approved for publication in the UK under section 21 of the FSMA by RFC Ambrian Limited (UK). It is being made available for distribution to
eligible counterparties and professional investors only (as those terms are defined by the rules of the Financial Conduct Authority (FCA)). Its contents are not
directed at retail clients. RFC Ambrian (UK) Limited publishes this document as non-independent research which is a financial promotion under FCA rules. It
has not been prepared in accordance with the regulatory rules relating to independent research and it is not subject to the prohibition on dealing ahead of the
dissemination of investment research.
RFC Ambrian Limited (ABN 59 009 153 888) is the holder of AFSL 233214.
RFC Ambrian Limited (Company number 4236075, registered in England and Wales) is authorised and regulated by the FCA.

8
RFC Ambrian Limited