The Investment Treaty Arbitration Review: Causation

I Introduction

The principle of causation is well established in national and international law.2 The International Law Commission Draft articles on the Responsibility of States for Internationally Wrongful Acts (the ILC Draft articles) state:3 'The responsible state is under an obligation to make full reparation for the injury caused by the internationally wrongful act.'

Demonstrating causation, therefore, is a prerequisite to reparation. The legal considerations for whether a claim demonstrates a sufficient causal link include the foreseeability, directness and proximity of the injury in relation to the alleged acts.4 A tribunal's evaluation of these considerations will depend on the specific facts of each case and the nature of the alleged acts.5

Separate from these legal considerations is factual causation,6 which 'mainly turns on proving, with evidence, that the particular unlawful act actually caused the damages in dispute'.7 There is no unanimously recognised standard of proof in international arbitration, although in practice the balance of probabilities standard is often applied.8

Factual causation sits at the intersection of disciplines between counsel and expert witnesses. While primarily a matter of fact, causation is often implicit to the scope of work of damages experts, who typically assess losses by reference to the breaches alleged by the claimants and, in doing so, will establish the link between a fact and a loss.

This chapter first discusses the interaction between causation, and the relevant framework to assess damages, before covering some of the ways in which experts can assist both claimants and respondent states with proving and disproving, respectively, factual causation.

II Interaction between causation and the relevant framework to assess damages

The full reparation framework aims to restore a claimant to the position in which it would have been, in all likelihood, absent the alleged injurious acts.9 Damages are correspondingly assessed by reference to the difference between the claimant's economic position in a hypothetical situation in which the injurious acts did not take place (known as the 'but for' or 'counterfactual' scenario), and the economic position in which the claimant actually finds itself (known as the 'actual' scenario).

However, the 'but for' analysis has its limitations:

A classic example is one where two hunters shoot at the same time at the victim who would have died by either of the shots. The but-for test holds that neither hunter caused the death since but-for his shot the victim would have died anyway by the shot of the other hunter.10

In the context of investment treaty arbitration, a commentary to the ILC Draft articles notes that 'international practice and the decisions of international decisions of international tribunals do not support the reduction or attenuation of reparation for concurrent causes, except in cases of contributory fault'.11 In other words, the state's actions do not have to be the only cause of the injury, but can be one of a number of causes with the state still being liable for full reparation.12

This approach would appear to lead to a just outcome for the claimant. If the victim had opportunities for recourse against both the state and the third party who concurrently caused the injury, any concerns regarding double recovery should be able to be addressed under the umbrella of mitigation.

An example of concurrent causes with contributory fault would be where there was only one hunter, and both the hunter and the victim shot the victim at the same time. Applied to an investment arbitration, this would logically result in damages being awarded in proportion to the harm caused and, therefore, would be borne equally between the claimant and the respondent state.

The actions of a third party can also lead to a finding of contributory fault. In Petrobart v. Kyrgyz Republic, KGM (a state-owned entity) owed money to the investor and the state was found to have breached its treaty obligations by removing assets from KGM. However, the tribunal found that KGM was in financial difficulties prior to the asset transfer and that, therefore, there was uncertainty as to whether the investor would have received its money in any event. As summarised by Ripinsky and Williams:

As a consequence of its finding that a concurrent cause of harm existed (ie, the pre-existing state of a third party) the Tribunal in 1 exercised its discretion to limit the amount of damages awarded to the investor. Emphasizing KGM's 'weak and unstable' financial situation even before the unlawful transfer, the Tribunal found that Petrobart, in any event, would not have been able to obtain full payment for its claim for delivered gas. As the Tribunal also recognized that it could not establish the precise value of the assets transferred from KGM, it proceeded to make a general assessment based on its appreciation of the situation as a whole. In making such an assessment the Tribunal found that the Kyrgyz Republic, as responsible for the transfer and lease of KGM's assets, must compensate 75 per cent of Petrobart's justified contract claims against KGM.13

The issue of contributory fault is not limited to consideration of concurrent causes, and tribunals have exercised their discretion in applying broad adjustments to damages claims upon a finding of contributory fault.14

Where the causes are not concurrent, the 'but for' framework can be applied. For example, in Biwater (BGT) v. Tanzania, the tribunal concluded that although 'BGT's investment . . . was the subject of an expropriation . . . by the time that this expropriation took place . . . the losses and damage for which BGT claims in these proceedings had already been (separately) caused'.15 The value of the claimant's investment was nil in the actual scenario and still would have been nil in the counterfactual scenario on the relevant valuation date.

The other exception set out in a commentary to the ILC Draft articles to the general principle that reparation should not be reduced for concurrent causes is in cases where 'some part of the injury can be shown to be severable in causal terms from that attributed to the responsible state'.16 To continue with the hunter analogy, if one hunter shot the victim in the arm and the second hunter shot the victim in the leg at the same time and the victim survived, the first hunter would only be liable for the injury to the arm. This is consistent with applying two 'but for' analyses, with the 'actual' situation reflecting the alleged fault of another party.

The language of the commentary to the ILC Draft articles appears to shift the burden of proof for demonstrating that some of the injury is severable to the state. This would suggest that where it has been demonstrated that the state's acts were 'a necessary element of a set of antecedent actual conditions that was sufficient for the causal result',17 the claimant is still entitled to full reparation from the state. However, some tribunals have instead 'focused on whether the conduct of the host State was the dominant or primary cause of the damage, especially where the factual matrix was complex and involved multiple causes'.18

In circumstances where the alleged injurious acts by the state did not take place at a single point in time, such as with creeping expropriation, it can be challenging to separate the effect of the acts from that of changes in market and macroeconomic conditions, or other changes in a claimant's fortunes.

One way to reduce uncertainty regarding causation is to be clear about the nature of the alleged injury and push any uncertainty regarding the extent of the injury to the assessment of quantum. Damages experts are well equipped to deal with uncertainty, which is inherent in estimating a counterfactual scenario.

For example, consider a claimant who was bidding for a concession to operate a casino, but whose gaming licence was illegally revoked by the state, thereby disqualifying the claimant from the tender process. If the nature of the alleged injury is the loss of opportunity to participate in the tender, then the causal link between the state's actions and the injury is clear. However, if the nature of the alleged injury is the loss of the concession contract, then the causal link is less clear. Assuming the claimant opts for the loss of opportunity claim, then uncertainty regarding who would have won the tender can be addressed through examination of the claimant's internal records indicating how their likely proposal would have ranked against the bid criteria. If no such records are available, or if the process itself was tainted by the state's actions, then the damages experts may give consideration to a pro rata chance of winning based on the number of qualifying bidders.

Given the wide-ranging approaches taken by tribunals when setting the threshold for causation and analysing the sufficiency of supporting evidence, claimants should be careful of (and respondent states should be wise to) bypassing uncertainties in the causal chain. Such behaviour may be driven by the perceived reluctance of tribunals to accept uncertainty in the damages calculation, despite such uncertainty being inherent in the 'but for' scenario.

III Delay analysis: the gold standard in causation analysis?

In construction disputes between owners and contractors, or contractors and sub-contractors, the causal link between liability and quantum is regularly scrutinised and debated between experts through what is known as delay analysis.

A typical construction claim involves an owner alleging that its contractor's failures have caused a project to be delayed, leading to losses being incurred. Construction contracts usually contain provisions for owner compensation in the event of delays, known as liquidated damages; for example, stating that the contractor will be fined a percentage of the contract value for each week of delay to completion, up to an agreed cap. These contractual provisions may be bypassed where material breaches of the contract (such as wilful default or misconduct) are alleged, leading to a more traditional loss of profits claim.

Regardless of whether the quantum is determined using liquidated damages or loss of profits, the delay analysis provides the causal link between liability and quantum. A common approach is to determine the contemporaneous critical path of the project in question (through the 'as built') and assess the extent and incidence of critical delay against the planned critical path of the project ('as planned'), attributing the causes of delay to the parties, other third parties or extraneous events such as extreme weather, by analysing contemporaneous records. This exercise is more complicated than it may seem, as contemporaneous records may incorrectly attribute perceived delays, particularly if prepared unilaterally.

An activity is determined as being on the 'critical path' of the project if a delay to the activity would result in a corresponding delay to project completion. The actual or contemporaneous critical path of a project, therefore, is dynamic as activities that were not planned to be on the critical path can become critical if sufficiently delayed. The assessment of the actual critical path, therefore, is carried at regular points in time during the life of the project.

Imagine a project to construct a hotel that was initially planned to take 20 weeks. The activity of installing access card readers to the rooms was initially planned to take place at the end of week 15 but was delayed by eight weeks. This activity was not on the planned critical path; the installation of the access card readers was not planned to affect or drive the initial completion duration of 20 weeks. However, construction progressed as planned and the access card readers were the only element delayed. This activity therefore moved onto the actual critical path and was only completed at the end of week 23, causing a three-week delay to the opening of the hotel. The critical path analysis allows for separable and proportionate damages to be assessed.

Critical path analysis, therefore, identifies the actual cause of delay, akin to identifying 'the dominant or primary cause of the damage'.19 Sometimes the critical path on a project can bifurcate, with two (or more) activities on the critical path at the same time. Where both activities are delayed, there would be concurrent causes of delay. If one activity was under the control of the owner and the other the contractor, how is the delay analysis affected? Guidance from The Society of Construction Law (SCL) states the following:

True concurrent delay is the occurrence of two or more delay events at the same time, one an Employer Risk Event, the other a Contractor Risk Event, and the effects of which are felt at the same time. For concurrent delay to exist, each of the Employer Risk Event and the Contractor Risk Event much be an effective cause of Delay to Completion (i.e. the delays must both affect the critical path). Where Contractor Delay to Completion occurs or has an effect concurrently with Employer Delay to Completion, the Contractor's concurrent delay should not reduce any [Extension of Time] due.20

In effect, where there is concurrent delay caused by both the owner and the contractor, the contractor is granted an extension of time (EOT) and, therefore, is not penalised for having caused any delay to the project. If the completion date is extended for reasons attributable to the owner, the contractor may be entitled to prolongation costs (such as the costs of having to remain on site longer). Does this mean that where there are concurrent causes of delay and the contractor is granted an EOT, the owner has to pay the contractor's prolongation costs? Here the SCL guidance again provides clarification:

Where Employer Delay to Completion and Contractor Delay to Completion are concurrent and, as a result of that delay the Contractor incurs additional costs, then the Contractor should only recover compensation if it is able to separate the additional costs caused by the Employer Delay from those caused by the Contractor Delay. If it would have incurred the additional costs in any event as a result of Contractor Delay, the Contractor will not be entitled to recover those additional costs.21

The SCL guidance in both instances reflects the application of the 'but for' analysis, with different perspectives framed by the alleged breach being tested. In the first instance regarding whether the contractor is liable for having caused the delay, as there were concurrent causes the project would have still been delayed but for the contractor's delay. In the second instance regarding whether the owner is liable for the contractor's prolongation costs absent the delay caused by the owner, the contractor would have likely still incurred prolongation costs and the onus is on the contractor to demonstrate otherwise. Neither party would be liable, therefore, for the other's losses or prolongation costs.

IV Approaches to analysing causation in non-construction disputes

Outside construction disputes, there is no single commonly accepted approach to demonstrating causation. Below we discuss some of the available approaches for analysing causation in non-construction disputes and give examples of the contexts in which they can be used.

i Benchmark analysis

Benchmark analysis seeks to demonstrate the effects of the alleged breach by comparing the abnormal performance of the investment or affected products following the breach to either (1) the performance of the same investment or affected products prior to the breach or (2) the performance of comparable investments or products during the breach period.22

Consider Company A that manufactures TVs using speakers supplied by Company B. Company A is claiming that Company B supplied defective speakers, which in turn led to a significant number of consumers returning the affected TVs, and Company A losing profits on the associated sales. The volume of sales is such that it is neither reasonable nor practicable to dismantle every returned TV to discover whether the speakers are faulty. However, Company A has done laboratory analysis on some of the returns to identify the affected components and models. Returns of TVs are commonplace for Company A, and are categorised by dealers generically (sound fault, picture fault, physical damage, etc.).

In this example, the impact of the defective speakers supplied by Company B could be determined by comparing the number of returns categorised under 'sound fault' for the affected models, which may also include sound faults because of faulty wiring, software or remote controls, to a benchmark level of 'sound fault' returns. The benchmark could be determined by reference to returns of unaffected models during the same period or returns of the affected models during an earlier period prior to the supply of the defective speakers. Taken in combination with the laboratory testing to find the root cause, the benchmark analysis would indicate that the excess level of returns of the affected models during the breach period were caused by Company B's supply of defective speakers.

This approach is most applicable in commercial disputes relating to high-volume products, as it relies on having sufficient data to define the benchmark. However, it can still be used in investment disputes in conjunction with an event study, as we discuss below.

ii Event studies

Event studies can be used for publicly traded companies to determine the effect on a company's share price and value of a given event or announcement, such as the implementation of a disputed government measure:

An event study is a well-established empirical technique that is used to measure the market's assessment of the effects of a specific even or announcement (here, for example, the disputed measure) on a company's market value. An advantage of the event study method is that it employs actual market data rather than potentially subjective assumptions.23

Take the example of a listed online retailer whose website in a given country was blocked by the state on 1 March, with both the retailer and the government issuing press releases on the same day. By 10 March, the share price of the online retailer had fallen by 20 per cent. If there were no other press releases or market data issued by the online retailer in the intervening period, then the event study may conclude that the government's actions caused the online retailer to lose 20 per cent of its equity value.24

However, share prices can be volatile, with many share price movements unable to be explained by a company releasing a positive or negative news story, and instead reflecting investor sentiment and wider market trends. As a result, a more nuanced approach is often required, and event studies will usually also give consideration to market movements in the observation window. This is akin to a benchmark analysis, with the purpose of the benchmark again being to isolate the abnormal performance related to the state's actions.

In our example, the movements in the relevant market could be determined by reference to the share price movements of comparable companies during the same period, or the movements in a representative index (such as an index of technology companies). If the share prices of comparable companies, unaffected by the alleged breach, went down by 5 per cent between 1 and 10 March, then the event study may conclude that only a 15 per cent fall in the online retailer's share price during the same period is attributable to the government's actions.

Regression analysis may be used on the benchmark to demonstrate the statistical significance of the claimant's excess decline, providing further comfort on the conclusion of the event study.25

iii Regression analysis

Regression analysis is a hidden feature of most damages claims; it is used in determining the discount rate applied in assessing the present value of future lost profits or cash flows. More specifically, the capital asset pricing model, commonly used to estimate a company's cost of equity, includes a component known as the 'beta', which measures the sensitivity of a given stock to market movements.26 The beta is calculated using a regression analysis of the stock's returns versus market returns.27

Another instance where regression analysis is often seen in damages claims is in the analysis of whether a cost category is fixed or variable for the purpose of estimating costs in the 'but for' situation; for example, by plotting the monthly costs in that category against production volumes.

Regression analysis can also be a useful tool for demonstrating causation, as in the example above of providing statistical confidence to the outcome of an event analysis. Furthermore, regression analysis can be used in any context in which the objective is to demonstrate that an outcome deviated from the expected result. Taking the example of the TVs with the faulty speakers, the number of returns categorised as 'sound fault' could be plotted against the sales volumes for each unaffected model to determine the expected returns for a given volume of sales. Adding the returns and sales volumes of the affected models could demonstrate that the number of returns was significantly above expectations.

The chapter on causation in the fifth edition of this publication addresses the use of regression analysis to assess causation in more detail and presents two hypothetical case studies.28

iv Cash flow analysis

Rather than being used solely for assessing damages, cash flow analysis can also be used to demonstrate causation.

Consider the example of a mining company claiming that its investment was expropriated as a result of the government assessing extra-contractual royalties. A year after paying these additional royalties, the company's mining licence was terminated because it failed to meet the minimum exploration expenditure conditions of its licence. Although the termination may have been valid, the assessment of the additional royalties was not. An analysis of the mining company's cash flows would reveal whether it would have been able to make the necessary exploration expenditure had it not paid the additional royalties the prior year, thereby establishing whether the royalty breach caused the licence termination.

Of course, this is a simplified example, and the reality for junior mining companies is more complex, and usually centres around balancing sufficient exploration, drilling and studies to demonstrate the attractiveness of the project to potential investors against the available funds. This, often delicate, balance can be upset by seemingly minor actions by the state, which can knock investor confidence and tip the scales against a junior mining company seeking to raise financing. The implications of such scenarios for damages will depend on the causation test applied by the tribunal and the contemporaneous factual record.

Other contexts in which cash flow analysis could be used to demonstrate causation include an investor going bankrupt, or breaching a loan covenant, following a state action that led to a cash outflow. Although not strictly a cash flow analysis, a similar approach could be used even where the state action led to a liability being recorded on the balance sheet, particularly in cases where this resulted in a loan covenant being breached.

v Decision tree analysis

Decision tree analysis involves consideration of whether the alleged injurious acts affected the decisions made by the investor with regard to their investment.

Imagine an investor in a solar plant who made the decision to proceed with its investment based on the feed-in tariffs published by the government at the time. The government later changed the tariffs to be much less favourable for the investor, and it transpired that the government had been planning to do so prior to the investor's original decision to invest but failed to disclose this fact despite being under a contractual obligation to do so. In such a case, the information concealed by the government could have resulted in the investor deciding not to invest, providing a causal link in support of a claim for its sunk costs.

The decision tree analysis could also be supported by a cash flow analysis; for example, showing that the returns on the project with the new tariffs would have been below the company's cost of capital, and the investment would therefore have been rejected by its investment committee. In other cases, such as where the concealed information removes the potential for any revenue generation, cash flow analysis may not be necessary.

V Conclusion

The requirement for a claimant to demonstrate a sufficient causal link between alleged breaches and an alleged injury is well established in international arbitration, and causation is an implicit step in the full reparation framework. However, causation is rarely sufficiently scrutinised outside construction claims, and often bypassed through the instructions given to experts.

The reality is that factors other than the state's alleged actions also contribute to an investor's fortunes (or change in fortunes), clouding the theoretical clarity of the full reparation framework. Whether and how a tribunal adjusts for these factors should depend on the causation test being applied and the facts of the case. In practice, tribunals often do not elucidate their reasoning. Nonetheless, uncertainty about causation may go some way to explaining why tribunals in investment treaty arbitrations are generally perceived to be permissive on liability and restrictive on quantum.

Part of the problem with demonstrating causation is that it often sits at the boundaries of legal, technical and damages expertise. Despite this, delay analysis in the context of construction claims provides a clear example of how to demonstrate the causal link between the actions of one party and the consequences for another, ensuring that the damages claim is appropriately framed. Although delay analysis methods cannot be applied directly to other types of claims, there are various other approaches that can be used individually, or in combination, to demonstrate causation.

Putting to one side the time and cost implications of carrying out a causation analysis, which can be significant in construction claims, there are two critical success factors that apply regardless of the type of claim and causation approach being applied. First, it is key to define with sufficient precision the actions (breaches) and consequences (injuries) that bookend the causation analysis. Without well-defined start and end points, the analysis risks being, at best, inaccurate, and at worst, irrelevant. Second, it is vital to have access to sufficient contemporaneous records to build a detailed factual chronology. This is often challenging in investment arbitrations owing to the passage of time and the misplacement, destruction or loss of custody of records. Document production often comes too late in the process, and tribunals can be reluctant to greenlight requests they perceive to be overly broad. This forces parties to narrow their requests, but it can be challenging to ask the right questions; it is all very well asking for the house with the red door, but not if you are standing in the wrong village.

Even if the causal link has been demonstrated, it is likely that some uncertainty will remain regarding the claimant's economic position in the counterfactual scenario. This is normal; uncertainty is inherent in any valuation exercise, and damages experts are well equipped to address it.

Despite its complexity, we are hopeful that the trend of causation attracting greater scrutiny in investment arbitration will continue and look forward to participating in related discussions in the arbitration community.


1 Anthony Theau-Laurent is a partner and Edmond Richards is a director at Accuracy. The authors would like to thank Juan Saez, Robert Deegan and Louis Hoghton for their assistance in reviewing this chapter.

2 Damages in International Arbitration (John A Trenor, ed., 4th edition, Global Arbitration Review), Chapter 5: 'Overview of Principles Reducing Damages', pp. 94–96.

3 International Law Commission, 'Draft articles on the Responsibility of States for Internationally Wrongful Acts, with commentaries' (2001) (ILC Draft articles), Article 31.

4 ILC Draft articles, Article 31, Commentary 10.

5 Ioannis Kardassopoulos and Ron Fuchs v. The Republic of Georgia, ICSID Case Nos. ARB/05/18 and ARB/07/15, Award (3 Mar. 2010), parags. 468–69. Sergey Ripinsky and Kevin Williams, Damages in International Investment Law, p. 136.

6 We note that the distinction between legal and factual causation may not always be clear. See Patrick W Pearsall and J Benton Heath, 'Causation and Injury in Investor-State Arbitration', Contemporary and Emerging Issues on the Law of Damages and Valuation in International Investment Arbitration (2018).

7 Damages in International Arbitration (op. cit. note 2, above), Chapter 5: 'Overview of Principles Reducing Damages', p. 95, quoting Borzu Sabahi, 'Compensation and Restitution in Investor-State Arbitration Principles and Practice', 171(2011).

8 ibid., Chapter 1: 'Compensatory Damages Principles in Civil and Common Law Jurisdictions: Requirements, Underlying Principles and Limits', p. 11.

9 ibid., Chapter 6: 'Full Compensation, Full Reparation and the But-For Premise', p. 111.

10 Ilias Plakokefalos, 'Causation in the Law of State Responsibility and the Problem of Overdetermination', European Journal of International Law 26 (2015), p. 477.

11 ILC Draft articles, Article 31, Commentary 12.

12 Ilias Plakokefalos (op. cit. note 10, above), p. 478.

13 Sergey Ripinsky and Kevin Williams, Damages in International Investment Law, pp. 144–45.

14 Sandart & Partners, 'Defences to Damages Claims under Investment Treaty Arbitration', Lexology (2020).

15 Biwater Gauff (Tanzania) Ltd. v. United Republic of Tanzania, Award (24 Jul. 2008), para. 485.

16 ILC Draft articles, Article 31, Commentary 13.

17 Richard W Wright, 'Causation, Responsibility, Risk, Probability, Naked Statistics, and Proof: Pruning the Bramble Bush by Clarifying the Concepts', 73 Iowa Law Review 1001, p. 1019 (1988).

18 Joachim Knoll and Tania Singla, 'Causation in International Investment Law, Putting Article 23.3 of the India Model BIT into Context', Indian Journal of Arbitration Law 8/2, p. 87 (2020).

19 See footnote 17.

20 The Society of Construction Law, 'Delay and Disruption Protocol' (2nd edition, 2017), 10.

21 ibid., 14.

22 Forecasts made prior to the breach could also be used, if it can be established that they are reasonable.

23 Damages in International Arbitration (op. cit. note 2, above), Chapter 28: 'Damages in Life Sciences Arbitrations', p. 420.

24 Disregarding investors' expectations of any recovery of the lost value through potential arbitration.

25 Damages in International Arbitration (op. cit. note 2, above), Chapter 28: 'Damages in Life Sciences Arbitrations', p. 423.

26 McKinsey & Company, Valuation: Measuring and Managing the Value of Companies (5th edition, John Wiley & Sons, June 2020), p. 239.

27 ibid., pp. 249–57.

28 The Investment Treaty Arbitration Review (5th edition, The Law Reviews), Chapter 24: Causation, pp. 300–05.

The Law Reviews content