## Abstract

The title ‘The Value of Outcrop Studies in Reducing Subsurface Uncertainty and Risk’ might suggest that we expect new information to improve prospect risk, but this is not correct. Gaining new information generally does change our estimate of prospect risk: the change may be up or down, and the average of all possibilities is zero change. You cannot acquire data for the purpose of increasing the probability of success. We should expect that: (a) the expected value of the risk of a single prospect, post-data, is equal to the prior (pre-data) value; and (b) risk should become worse for the majority of prospects. New information adds value, not from changing pre-drill risk, but from decisions made as a consequence. The main value added is from identifying prospects not to drill, thereby saving the cost of likely dry holes, and by choosing the ones to test first, accelerating revenue from a successful outcome. Further value is added if the new information leads to the identification of new prospects or new plays, or suggests follow-on potential elsewhere in the region. Value may also be added if the new information is negative for a whole region, enabling us to focus our attention elsewhere.

Readers with experience in the petroleum industry may have heard statements along the lines of ‘we will acquire 3D data and this will reduce our prospect risk’, or ‘your task is to work up these three prospects, improve their risk, and promote them to drillable status’. Implicit to such statements is a belief that it is possible to design a work programme that will change our estimated chance of prospect success for the better. This article explores the possible root of this belief in the Labour Theory of Value, and shows that it is not valid when applied to prospect risks.

## Why we believe that doing work adds value: and why this does not apply to prospect risk

The Labour Theory of Value (Smith 1776) is founded on the observation that effort commonly results in increase of value. This concept influences philosophy, economics and, to some extent, hydrocarbon exploration. Bertrand Russell (1946) traced its origin to the statement by Thomas Aquinas (1265) that ‘value can, does and should increase in relation to the amount of labor which has been expended in the improvement of commodities’. This was key to the economic philosophy of Adam Smith and others; Karl Marx (1898) went further, stating ‘the value of a commodity is determined by the quantity of labour bestowed upon its production’.

The value we assign to an undrilled prospect portfolio is dependent on many factors, which include the expected number of discoveries and the risked volume of hydrocarbons we believe they contain. Thus the perceived value of a portfolio depends critically on the chance of success we assign to the prospects within it. Applying the Labour Theory of Value to the business of oil and gas exploration gives rise to a belief that is common, seductive and wrong; namely that by acquiring information (whether in the form of seismic data acquisition, analogue field studies or whatever) we expect to increase the value of a prospect, or a portfolio of prospects, by improving the risk (geological chance of success). Indeed, the title of the 2014 Geological Society meeting ‘Reducing Subsurface Uncertainty and Risk through Field-based Studies’ (Geological Society 2014) can be read as an expression of this expectation – if we carry out studies (in this case field-based studies) we expect to reduce subsurface risk.

This belief can have a severe impact on companies and on individuals. We have seen cases where performance contracts require a team to improve the chance of success of key prospects by remapping and further geological studies; companies have committed to 3D seismic programmes with the expectation that the data will lead to improved chance of success of existing prospects.

The Labour Theory of Value appeals to the human desire that our efforts are worthwhile. The contrary view is that ‘all the work one cares to add will not turn a mud pie into an apple tart; it remains a mud pie, value zero’ (Heinlein 1959). Our problem in exploration is that we do not know whether the undrilled prospect is a figurative apple tart or a mud pie; all we have is an estimate of the relative likelihood of the good outcome v. the bad outcome, based on our current state of knowledge and the information available to us at the time. This article explores the implications that acquiring new information has on subsurface (prospect) risk.

## The meaning of ‘expectation’

How we use words in common language can be very different from how we use the same words in formal science, and this can cause confusion when we use these words to discuss exploration risk. In common language, we ‘expect’ an outcome that commonly occurs, such as the birth of a child (Murkoff & Mazel 1984). It has a different meaning in probability and statistics, where the expectation is the predicted average outcome of an experiment (or process). The expected value is the sum of ((each of the possible outcomes)×(the probability of the outcome occurring)). The expected value may not occur frequently, or may indeed never occur.

The expected number of children in a family in the UK is 1.7 (UK Office of National statistics, 2012) but even a statistician would be surprised to have 1.7 children. The expected value (in the formal sense) of a fair, six-sided, dice is 3.5, being the average of all outcomes (1+2+3+4+5+6)/6: to actually land a score of 3.5 on one toss would be most unexpected (in the informal sense).

In petroleum exploration, we commonly use the word in its formal sense, for example EV (Expected Value) is the probability-weighted average value of all identified possible outcomes, and we will use the words ‘expectation’ and ‘expected value’ in that sense throughout this article.

## The Law of Total Probability and its impact on prospect risk

The Law of Total Probability is a fundamental part of modern probability theory and it is of vital significance to this discussion. Salkind (2010) noted that the law is implicit to Bayes's Theorem (Bayes 1764). The Law is widely stated in modern texts (e.g. Bean 2001; Cyganowski *et al.* 2002; Salkind 2010) without an original attribution, but it can traced back to the Theory of Inverse Probability (Laplace 1774). When applied to the business of oil and gas exploration, it creates some non-intuitive results, which run counter to the Labour Theory of Value. Let us consider the case of an oil prospect, based on 2D seismic data, for which we have a prior chance of success, and over which we are acquiring 3D seismic data. This is illustrated graphically in Figure 1a.

The Law of Total Probability can be expressed as
where *P*(*B*) is our prior probability of an outcome *B* (such as an exploration well success); this is the initial estimate made prior to gaining new information. *A* represents some event (such as acquiring and interpreting new 3D seismic data) that will provide new information that may cause us to revise that prior estimate. That event *A* has *n* possible outcomes, such as:

*A*_{1}=3D confirms closure exists, and shows amplitudes that conform to closure;*A*_{2}=3D confirms closure but has no amplitudes;*A*_{3}=3D confirms closure does not exist, etc.

*P*(*Ai*) represents the likelihood of each of those outcomes happening and *P*(*B*|*Ai*) is the revised probability of outcome *B* (the new estimate of chance of success, made using the new information).

The Law of Total Probability states that the mean outcome – the expected value of the chance of success after we have the new information – is equivalent to the prior probability *P*(*B*), estimated prior to obtaining that information. In other words, in our example, the expected value of the prospect risk after we have acquired 3D data should be exactly the same as the prior; some outcomes changed the risk for the better, some for the worse, but the mean outcome is no change. This can be expressed by the adage ‘you can't put oil in the ground by shooting seismic data’.

This law can be generalized to describe a situation in which the new information does not deliver *n* discrete possible outcomes, but instead delivers a continuum of possible results, represented by a continuous distribution; again the mean of the outcomes, weighted according the probability of each outcome, is equivalent to the prior chance.

A simple case (Fig. 1b) in which the information can only deliver two possible outcomes – good news or bad news – gives a simpler equation:

In effect, this law means that the expected outcome of new information is no change. We should not expect that acquiring new information should improve our chance of success: it might do so, or might not. We emphasize that we are using the word ‘expected’ in its formal sense, as described above; in informal language, we will be surprised at any outcome where value does not change, but the average of all possible outcomes is no change.

Because this law can appear quite counter-intuitive, and because it is so important to our case, we will illustrate this by two simple examples of binary outcomes, in which we provide the numerical values and demonstrate the validity of the law in these cases.

In the simple game of dice shown in Figure 2, we win if the dice are both showing one spot. Today, the dice have been cast, but they are both covered, and we do not know which way they have fallen. Today, we estimate the probability of winning as 1 in 36; this is called the prior probability. If winning has a value of $100, and there is no cost to losing – our expected value of the game is $100/36=$2.78.

Tomorrow, we will obtain better information: one of the cups will be removed. Our estimate of the chance of success will change, depending on what we see. This new chance of success is called the conditional probability (Popper 1959; Hájek 2003). It may be better, or it may be worse, than today's estimate; if the exposed dice has one spot, we are still in the game, and only the second dice has to have the winning number; the new probability will be 1/6. From today's perspective, we describe this as the conditional probability if the exposed dice shows one spot. However, if the exposed dice has more than one spot, there is no way we can win, and the new probability will be zero. In this simple game, we know exactly what the chance of change in either direction is; there are five ways it can be bad news (anything but one spot showing) and one way it can be good news. The conditional probability will improve to 1/6 in 1/6 of the outcomes, and it will get worse, decreasing to zero, in 5/6 of the outcomes. The expected value of the conditional probability is the probability-weighted average of these, (5/6×zero)+1/6×1/6)=1/36. The expected value of the game, from today's standpoint, remains $100/36=$2.78.

Note that this value is unchanged from today's estimate of the chance of success: the Total Probability Rule, or Average Rule, which derives from Law of Total Probability, effectively states that the probability weighted average of possible outcomes (in this case the conditional probabilities) is equal to the prior expectation (in this case the prior probability).

We emphasize that, in each possible outcome in this game of dice, gaining new data does change the probability of success, but you cannot know beforehand whether the probability of success will go up or go down. Thus you cannot acquire data for the purpose of increasing the probability of success. In the case of the simple game of dice shown here, the application of Law of Total Probability is intuitively obvious: in more complex scenarios, typical of petroleum exploration, the same rule applies, but it is less intuitively obvious.

Exactly the same principle applies in the game of petroleum exploration, as shown in Figure 3. We can consider a prospect for which today's state of knowledge is limited to a sparse 2D grid. Contouring the depth values seen on the 2D lines indicates the possibility of a closure, and we estimate that the chance of this trap geometry existing is 0.4. Our expected value of the game is 0.4 of a discovery. Tomorrow, we will receive a 3D dataset that we think will either confirm or deny the existence of the closure. We should expect the average probability after receiving the new information to be the same as the prior probability; from today's perspective, the expected value of the game is still 0.4 of a discovery. In other words, our new estimate of the chance of success of the prospect may become better or worse, but the average of all possible outcomes is unchanged.

In these two games, the result of the new information was binary (there were only two, discrete outcomes). The same principle applies when the new information can give many different outcomes, or the possible outcomes form a continuous distribution: the expected value of the conditional probability is the same as today's prior probability.

## What is prospect risk?

Prospect risk estimation is a vital part of the management of an exploration portfolio (e.g. Allais 1956; Newendorp 1975; Megill 1977; Rose 1987). Prospect risk (also known as *geological risk*, or the *chance of geological success* of a prospect) describes our current opinion of the likelihood that the success case geological model is correct, and that the parameter ranges associated with that success case scenario are appropriate. This estimate is based on our current understanding and experience and the information currently available to us. It is, therefore, not an attribute of the geology of the prospect; instead it is attribute of the observer. The estimate will change with time, dependent on the information available about the prospect, the level of understanding about the geological model, and to some extent on the world view of the observer.

‘Prospect risk’ commonly refers to the chance of success, but it has also been used to mean the chance of failure, generating confusion as to whether ‘high risk’ means likely failure or likely success. To avoid this confusion, it is advisable to avoid the term ‘risk’ other than for informal purposes, and refer specifically to the chance of geological success. There is no consistent terminology for this concept. This article follows Rose (1987, 1992, 2001) in using the symbol *Pg* (Probability of Geological Success). Abbreviations used by other sources for the same quantity include GP (Geological Probability; Ross 1997, 2004), GPoS (Geological Probability of Success), CoS (Chance of Success) and POSg (Quirk & Ruthrauff 2008).

In this article, we quantify the chance of success as numerical probability (0–1, where 0 is no chance of success and 1 is no chance of failure). Other publications quantify it in the form of odds (e.g. 1:10) or as a percentage (0–100%).

## How does prospect risk change in the light of new information?

Applying the principles described above to the process of oil and gas exploration, we can derive two important rules for how the prospect chance of success will change, both of which are counter-intuitive, and both of which may initially appear to undermine our rationale for acquiring new information. However, by recognizing that we add value to the exploration cycle by making good decisions (Bratvold & Begg 2008, 2010), not by changing prospect risks, we can demonstrate how new information adds value (Howard 1966).

### Rule (i) The expected value of *Pg* will be unchanged (i.e. you shouldn't expect to improve *Pg* by acquiring new data)

If we acquire new information, it helps us make a more reliable estimate of that probability; it may bring good news or bad news, so that our subsequent re-evaluation of the chance of success (i.e. our conditional probability, given the new information) may change for the better or for the worse. The magnitude and direction of the change depend on our prior estimate, on whether the new information appears positive or negative, and how reliable the new information is (Bayes 1764; Laplace 1774; Morgan 1968; for specific application of this method to petroleum exploration, see for example Newendorp 1971, 1975).

Although we expect that our *Pg* estimate for individual prospects will, in most cases, change in the light of new information, we do not know the direction of the change. As described above, the law of Total Probability dictates that the mean value of all the possible outcomes is unchanged.

In other words, we should not expect to reduce risk (i.e. increase the chance of success, *Pg*) by acquiring new information, although we may hope that this is the result. This is illustrated schematically by Figure 4 which shows a prospect for which the present-day estimated *Pg* is 0.25. After acquiring new information, the new estimate of *Pg* is likely to change, falling within a range of possibilities: but the average of all these possibilities is the same as the prior estimate, 0.25.

### Rule (ii): you should expect *Pg* to become worse for the majority of prospects

The majority of real-world exploration prospects have pre-drill chance of success significantly less than 0.5. Quirk & Ruthrauff (2008) showed the average chance of success of exploration wells in 18 oil and gas plays, ranging from 0.2 to 0.48, with a mean of *c.* 0.35.

With additional information, some prospects will see improved *Pg*, and some will see diminished *Pg*, but if the prior *Pg* is below 0.5, the upgraded ones can move further in the positive direction than the downgraded ones can move in the negative direction (see Fig. 5). Therefore, because the expected value of the average prospect risk of all possible outcomes should be unchanged, we should expect that more outcomes will be downgraded than are upgraded.

This is illustrated graphically in Figure 5, which shows the range of change in *Pg* that might result from gaining a particular piece of information whose outcome is a distribution. The new *Pg* estimate is represented by a probability density function (pdf) curve (Ushakov 2001). A typical distribution expected for a prospect with *Pg*<0.5, is skewed, with the median value smaller than the mean.

Some types of new information are likely to result in a simple skewed distribution, as shown in Figure 5. Other types of new information, which are more diagnostic of success v. failure (e.g. seismic attribute data in a setting where hydrocarbon anomalies are expected to have a strong seismic response), are likely to result in more polarization and a bimodal distribution (Fig. 6).

For our notional prospect, which has a prior (pre-information) *Pg* estimate of 0.25, we would expect that information will improve our *Pg* estimate for about a quarter of the future scenarios, and it will make it worse in about three-quarters of the future scenarios (Fig. 4). The numerical ratio of gainers v. losers in these future scenarios depends on the nature of the new data and the way and the degree to which it can polarize our estimation of *Pg*.

The basis for this is explored in more detail in Peel & Brooks (2015), who use a probabilistic forward model of prospect risk estimation to demonstrate the evolution of *Pg* for a prospect with initial *Pg* of 0.25 through four successive stages of knowledge/data gain. Their model shows that the ratio of gainers (improved *Pg*) to losers (decreased *Pg*) does tend to give more losers than gainers for a prospect with initial *Pg*<0.5, but that this ratio depends strongly on the nature of the new information. Some types of information may have the opposite effect (more gainers than losers). Their modelling did, however, in all cases support the applicability of the law of Total Probability to prospect risk.

The counter-intuitive implication of this is that, if we have an original estimate of the chance of success that was appropriately estimated, and we acquire appropriate data, and we work on it correctly, the expected result will be that the majority of exploration prospects should become riskier (lower *Pg*) with time, and that risk will only improve for a minority of prospects. This outcome does not mean that the original *Pg* estimate was in any way optimistic.

The perceived *Pg* (chance of success) of this prospect may continue to evolve as our state of knowledge and data increase (Fig. 6), but we do not know in advance whether *Pg* will improve or become worse. The distribution of possible outcomes is expressed here as a probability density function (pdf). As more information becomes available, the pdf curve, which represents future outcomes of where *Pg* may move to, as seen and predicted from our current perspective, broadens and changes shape, representing the effect of possible positive or negative evidence that we will gain in the future. Information obtained by fieldwork may provide evidence of some or all of the critical geological elements working regionally (or of them not working). On the schematic example shown in Figure 6, the influence of field studies is shown as moving the pdf curve a short distance up its evolutionary track.

Given enough constraint and suitable geology, the perceived *Pg* may become polarized, giving a bimodal pdf curve, in which the prospect either appears to have very strong evidence in favour of success, or very strong evidence against success. This generally is only achieved with high-quality 3D seismic reflection data containing apparent hydrocarbon indicators (such as AVA anomalies, depth-conforming amplitude anomalies, flat spots, etc.)

We emphasize that the broadening pdf curve shown in Figure 6 represents our uncertainty, from today's perspective, about where we will be in the future, and that our uncertainty about the future estimate increases as we look further down the path. This does not mean that our uncertainty will increase in the future; in each of those future worlds, we will have a single estimate, and our confidence that the estimate is appropriate should be higher.

An exception to Rule (ii) is that, for the minority of prospects for which our prior estimate of *Pg* is higher than 0.5, we should expect that obtaining new information will improve the *Pg* on more prospects than are downgraded, but as before, the expected value of the revised *Pg* is unchanged.

In every case, there is an expectation, in the informal sense, that the estimated chance of success will change from its current (prior) value in the light of new information. We do not know whether this change will be upwards or downwards. The expected (in the formal sense) posterior value of the new, revised estimate made using the new data is equal to the prior value.

## Application to a portfolio of prospects

The same logic applies to a portfolio of prospects. Figure 7 shows a map of a hypothetical lease block containing 10 mapped prospects. At present day, each of these has an estimated *Pg* of 0.2, and the expected number of successes if all are drilled is 10×0.2=2.0; if each prospect has a mean success case volume of 100 mmbbl, we can estimate given the information available today that the total risked volume of hydrocarbons on the block is 200 mmbbl. After acquiring new information, our perception of risk may change for the better or for the worse, but the mean of all these possible future scenarios is an expected number of successes which is unchanged at 2.0, and an unchanged total risked volume.

Again, with prospect sets as with individual prospects, if prior *Pg*<0.5, we expect the information to be disappointing more often than it is encouraging, but the fewer good cases should improve more than the many disappointing cases, so the net effect, on average, is no change. The majority of exploration blocks should become riskier with time, and risk will only improve for a minority of blocks. This outcome does not mean that the original risk was in any way optimistic.

## Shared risk, play risk and dependency

In the preceding analysis, we have, for the sake of simplicity, considered the *Pg* of each prospect to be independent from that of the others. It is common for sets of prospects to carry shared risk elements, such that they are not fully independent. If the prospects fall within an unproven play (for a review, see White 1993), we can assign a shared play risk component *P*(play), such that the chance of success of the individual prospect, *Pg*=*P*(play) **P*(prospect|play). This is an effective way of describing and analysing prospect portfolios, but it does not permit the prospect analysis to breach the Law of Total Probability. As before, new information can prove the play, changing *P*(play) to 1, or prove it invalid, changing *P*(play) to zero, or it can change our estimate of *P*(play) to a new intermediate value. However, the expected value of the conditional probability (i.e. the new estimate, given the new information) is equal to the prior estimate; in some cases, it may improve, in some cases, it may worsen, but we should not expect that by acquiring new information we can improve the chance of success (rule i) and if the play risk is below 0.5, we should expect more outcomes in which it worsens than outcomes in which it improves (rule ii).

## If we do not expect the information to improve *Pg*, how does it add value?

Risked volume is a philosophical concept with no concrete existence. Changing the perceived risk on a prospect only has a physical expression in value terms if it causes some action which involves cost or revenue. The principal value effect of refining our pre-drill estimate of *Pg* derives from the money saved by NOT drilling wells. Identifying and avoiding the prospects with least chance of success, and thereby avoiding the sunk cost of a dry hole, is a genuine value addition. Conversely, improving the perceived *Pg* of a prospect which was already on the drilling schedule adds no material value.

The way in which new information adds full-cycle value is described by Value of Information Theory. There is extensive literature on this subject, for example Howard (1966), Howard & Matheson (1984), Coopersmith & Cunningham (2002), Coopersmith *et al.* (2006); for a review of the literature, see Bratvold *et al.* (2009) and Bratvold & Begg (2008, 2010).

This is illustrated by the hypothetical lease block shown in Figure 7, which contains 10 prospects, all the same size, with initial *Pg* of 0.2. If we take the cost of an exploration well to be $100 MM and the value of a discovery to be $500 MM, the most likely outcome of drilling out the 10-prospect portfolio is two discoveries. This gives a value of $1000 MM for the outlay of $1000 MM on 10 exploration wells, giving zero net profit.

As shown in Figure 8, we may obtain new information (possibly including field data) at a cost of $100 MM, which enables us to refine the prospect risk estimates as shown. This refinement has in this case resulted in no net change in total risked volume, and the expected result of drilling out the portfolio still adds up to two discoveries.

The value of obtaining the new information derives from the wells that we choose not to drill. Figure 9 shows the predicted risked value of drilling through the portfolio, starting with the best (highest *Pg*) and working down the list, but stopping at different stages. In this case, value is maximized by choosing to drill only four wells, giving an expected net monetary value of +$235 MM, consisting of an expected value of 1.47 risked discoveries, worth $735 MM, for the cost of four wells ($400 MM) and the cost of obtaining the new information ($100 MM). (The expected value is based on the expected number of discoveries, which is simply the sum of *Pg* of the drilled prospects. The process of exploration drilling will deliver a real number not the expected number – the result of an exploration well is either a discovery or a dry hole. The most likely actual outcomes of this four-well campaign are either one discovery , giving a net value of zero, or two discoveries, giving a net value of +$500 MM, with these two outcomes about equally weighted.)

In this case, if we used the information to optimize the drilling campaign, the information has added a value of $235 MM. The key to realizing this value lies in knowing when to stop drilling.

If there is no flexibility in the number of wells to be drilled (e.g. if all 10 are obligatory commitment wells), the information did not, at face value, make any positive contribution. Yet even in this circumstance, some value may still be added from the new information through the time value of money (Capen 1995) by changing the drill order to promote the prospects with highest chance of success, and delaying those with lowest chance of success.

Additional exploration efficiency, and thus value, can also be achieved if we use the principles of play analysis set out by White (1993). We do not acquire new information expecting it to improve our play risk (the expected value is no change). We hope for an improvement in the play risk, ideally changing its status to proven, but if we already planned to drill out the portfolio, the upgrade in the play risk does not lead to added value. Information which allows us to disprove a play has more value impact because it allows us to choose not to drill any of the wells dependent on the play.

## Additional sources of value added by new information

The purpose of this article is to show how increased information should not be expected to change our estimate of the average risk on existing prospects, and to show that, for the majority of prospects (those with *Pg*<0.5), new information will lead to a reduction in our estimated chance of success.

While we believe that these conclusions are important, and that they should inform our decisions, we do not wish to suggest that this is the only reason for acquiring new information, nor that the only source of value lies in the existing prospect portfolio.

In addition to any value change that may occur as a result of changing the *Pg* (risk) estimate for a suite of existing prospects, as discussed above, new information such as that provided by field studies may add substantial value in other ways. These include the following:

identification of additional prospects in the licensed area, within the same geological play;

identification of new plays within the same area;

identification of new opportunities within the region (e.g. follow-on potential in adjacent licence blocks);

Even if the results of the new information are strongly negative, the learnings may result in a positive change relative to the uninformed situation, for example by providing the evidence needed to divest or to farm down the opportunity set, or simply to cut losses and focus attention and resources elsewhere.

## Conclusions

There is an expectation (in the informal sense) that the estimated chance of success of a prospect will change from its currently estimated (prior) value in the light of new information. We do not know whether this change will be upwards or downwards. The Law of Total Probability dictates that the expected (in the formal sense) posterior value of the new, revised estimate made using the new data is equal to the prior value. Thus we cannot carry out studies or acquire data for the purpose of increasing probability of success. This does not mean that there is no value in acquiring that new information: on the contrary, it may be vital that we do acquire the new information, because it can be the key to maximizing the value of our overall exploration programme. In some outcomes, we may indeed be rewarded by an improved chance of success.

We do not expect to reduce subsurface risk (i.e. to increase *Pg*) for a given prospect, or for a group of prospects, by carrying out field-based studies, or for that matter, by carrying out any kind of study. In some cases, this desirable improvement may come to pass, but in other cases the risk will worsen; the expected change is, on average, zero.

Given that the typical exploration prospect has *Pg*<0.5, we expect that more prospects will worsen than will improve, but for the minority of prospects that gain, the positive *Pg* change is likely to be greater than the negative *Pg* change for those that worsen. The net effect is zero average change.

For an existing portfolio of prospects, as with an individual prospect, we expect that the mean result of new information is zero net change in aggregate risk; again we expect numerically that more prospects sets will worsen than improve, but the few winners gain by a larger margin, giving zero net change.

Given improved knowledge to discriminate which prospects are more attractive than others, we are empowered to be selective regarding which subset to drill and in which order to drill them, which can favourably impact the overall economics of a drilling programme.

Obtaining new information adds value to an exploration programme mostly by saving money on the wells we choose not to drill (because their *Pg* has worsened); improving the *Pg* on a well that would be drilled anyway does not add real value. Therefore, ironically, obtaining bad news (for the prospect) before it is drilled is good news (for the company) because it helps us to avoid drilling costly dry holes. Value is also added by accelerating the testing of more favoured prospects for which *Pg* has increased.

New information can be expected to add significant additional value, not by reducing risk on existing prospects, but by identifying new opportunities (new prospects, new plays and new areas into which existing plays can be extended).

## Acknowledgments

The authors gratefully acknowledge support for Frank Peel by NERC, the UK Natural Environment Research Council. We recognize the incisive reviews provided by Peter Carragher and an anonymous reviewer which contributed greatly. We are grateful to David Labonte who provided us with modern references to the Law of Total Probability.

- © 2016 The Author(s). Published by The Geological Society of London. All rights reserved