Design and Performance Aspects of Information Security Prediction
Markets for Risk Management
Pankaj Pandey and Einar Arthur Snekkenes
Norwegian Information Security Lab., Gjovik University College, Teknologivn. 22, 2815 Gjovik, Norway
Keywords:
Information Security, Security Economics, Security Risk Management, Prediction Market.
Abstract:
Prediction Markets are the markets designed and operated to mine and aggregate the information scattered
among the traders. Recently, some researchers have started exploring the application of prediction markets in
the information security domain. The information security prediction market will facilitate trading of contracts
to hedge the financial impact of the risks associated with the underlying information security events, such as
discovery of a vulnerability in a piece of software. However, prediction markets differ in their objectives and
requirements, and therefore information security prediction markets need to be carefully engineered to meet
the specific requirements. The contribution of this paper is the identification of a set of design requirements
for an information security prediction market, and associated performance criteria. We present five categories
of design requirements: Contracts, Trading Process, Participants and Incentives, Clearing House, and Mar-
ket Management for the information security prediction market. Furthermore, we present six performance
measures: Information Elicitation, Transparency, Efficiency, Transaction Cost, Liquidity, and Manipulation
Resistance for the performance assessment of information security prediction market.
1 INTRODUCTION
The Global Risk report of World Economic Forum
states that ”Effective methods for measuring and pric-
ing cyber risks may even lead to new market-based
risk management structures which would help in un-
derstanding the systemic interdependencies in the
multiple domain that now depend on cyberspace”
(WEF, 2014). A carefully designed Prediction Market
can be used as a market mechanism for the manage-
ment of information security risks.
Prediction markets are the markets designed and
operated for mining and aggregation of information
which is scattered among the market participants
(Berg and Rietz, 2003). Subsequently this informa-
tion is reflected in the market prices and the prices
have a strong correlation with the probability belief of
the traders (Luckner, 2008). Prediction markets have
been used for various purposes such as prediction of
government policy actions, weather events, economic
indicators, elections, etc. (Luckner, 2008). Prediction
markets have also been attempted in the security and
terrorism domain. A well known (now abandoned)
project in the security domain is ”Future Markets Ap-
plied to Prediction (FutureMAP)” project of Defense
Advanced Research Project Agency (DARPA), USA
(Hanson, 2006). FutureMAP was to be used as an
”Electronic Market-Based Decision Support” system
to improve the approaches for collection of intelli-
gence information (Hanson, 2006).
(Pandey and Snekkenes, 2014a) assessed the ap-
plicability of prediction markets inthe information se-
curity domain as a risk management tool in an intra-
organizational setting as well as in an open setting.
The usefulness of prediction markets in hedging the
(financial impact of)information security risks is ex-
plained in (Pandey and Snekkenes, 2014b). As the
prediction markets differ in their objectives and re-
quirements, an important question in the context of in-
formation security prediction markets (ISPM) is: how
to design and engineer an ISPM to achieve the ob-
jectives of information aggregation and risk manage-
ment? The primary contribution of this article is the
identification of a set of design elements and associ-
ated performance measures for ISPM.
The remainder of the paper is structured as: Sec-
tion 2 explains the research method followed for the
article. Section 3 presents the related work. Sec-
tion 4 presents the design requirements for an ISPM.
Section 5 explains the design elements of an ISPM.
Section 6 presents the performance measures for an
ISPM. Section 7 resents the conclusion and directions
for future work.
273
Pandey P. and Arthur Snekkenes E..
Design and Performance Aspects of Information Security Prediction Markets for Risk Management.
DOI: 10.5220/0005547502730284
In Proceedings of the 12th International Conference on Security and Cryptography (SECRYPT-2015), pages 273-284
ISBN: 978-989-758-117-5
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
2 RESEARCH METHOD
The research follows the Design Science Research
Approach (DSRA). DSRA is useful when innovations
and ideas are created for development of technical ca-
pabilities and products which will be instrumental in
effective and efficient process development for arte-
facts (Johannesson and Perjons, 2014). The steps in
DSRA are as follows:
1. Explicate Problem : The first step is to formulate
the initial problem, justify its importance and in-
vestigate the underlying causes (Johannesson and
Perjons, 2014). To explicate the problem we
started with examining the literature on informa-
tion security markets, and taxonomy of prediction
markets to design an ISPM. This enabled us in
identifying the limitations of existing market en-
gineering frameworks. An overview of reviewed
literature is covered in Section 3.
2. Define Requirements : The second step is to iden-
tify and outline an artefact to address the expli-
cated problem and to elicit requirements for the
artefact (Johannesson and Perjons, 2014). A re-
quirement is the property of the artefact that is de-
sired by stakeholders in practice and is used for
design and development of the artefact. A require-
ment can be functional, structural, or environmen-
tal in nature. The requirements for the artefact
(ISPM) are given in Section 4.
3. Design, Development and Demonstration of the
Artefact : The next step leads to creation and
demonstration of an artefact that fulfils the re-
quirements identified in the previous (second)
step. This includes designing the functionality
and structure of the artefact (Johannesson and Per-
jons, 2014). The demonstration shows that the
artefact can, in fact, solve the problem (or some
aspects of it) in the given situation. The func-
tionality and structure of the artefact (ISPM) are
demonstrated in Section 5.
4. Evaluation of the Artefact : The last step is to
evaluate the artefact. This determines the extent
to which the artefact is able to solve the explicated
problem and its requirements (Johannesson and
Perjons, 2014). We have not evaluated the artefact
(ISPM) in the strictest sense; however we have
identified the performance factors for ISPM in
Section 6. We have used the ’informed argument’
form of evaluation. In this form, researchers eval-
uate the artefact by reasoning and arguments for
its usefulness in meeting the defined requirements
and solving the explicated problem. Informed ar-
gument form of evaluation is often used to evalu-
ate the artefacts which are highly innovative and
are still immature.
3 RELATED WORK
(Weinhardt and Gimpel, 2007) defined a market as ”a
set of humanly devised rules that structure the inter-
action and exchange of information by self-interested
participants in order to carry out exchange transac-
tions at a relatively low cost”. The efficiency and ef-
fectiveness of the market in achieving the specific ob-
jectives depends upon the design and implementation
of the market.
(Spann, 2002) proposed a taxonomy for the imple-
mentation of (prediction) markets. This taxonomyhas
ve elements with several sub-components, namely
Market Strategy, Market Design, Information Design,
Market Operationsand Data Interpretation. Out of the
ve elements only one element, i.e. Market Design is
relevant for this article. The ’Market Design’ com-
prises of six elements: Underlying (event), Medium
of Exchange, Incentive System, Trading Mechanism,
Market Rules and Kick Off Settings.
(Weinhardt and Gimpel, 2007) proposed a ’Mar-
ket Engineering Framework’ to define a structured,
systematic and theoretically grounded process of de-
sign, implementation, evaluation and introduction of
market platforms. (Plott and Chen, 2002; Luckner,
2008; Sripawatakul and Sutivong, 2010) also present
some guidelines on the design and implementation of
prediction markets.
However, due to various legal, intellectual prop-
erty and security reasons (Fidler, 2014), the design
and implementation issues discussed in the above arti-
cles do not suffice to address the specific requirements
and objectives of the ISPM. Further, we need specific
parameters to understand and assess the performance
of an ISPM.
4 REQUIREMENTS FOR ISPM
From our systematic review of literature on design
and implementation of prediction markets (Luckner,
2008; Weinhardt and Gimpel, 2007; Spann, 2002;
Plott and Chen, 2002; Sripawatakul and Sutivong,
2010), and existing market methods for management
of information security risks (Fidler, 2014) we have
identified 13 key design requirements for an ISPM.
As shown in Table 1, the design requirements are
grouped into five categories: (i) Contracts, (ii) Trad-
ing Process, (iii) Participants and Incentives, (iv)
Clearing House, and (v) Market Management.
SECRYPT2015-InternationalConferenceonSecurityandCryptography
274
Table 1: Requirements for ISPM.
Contracts
Type of Contracts
Contract Specifications
Trading
Process
Trading Mechanism
Anonymous Trading
Participants
and Incentives
Participant’s Motivation
Incentive Structure
Clearing
House
Contract Settlement Criteria
Counterparty Risk Management
Trusted Third Party
Intellectual Property Management
Know Your Trader
Market
Management
Regulated Market
Legal Permission
5 DESIGN ASPECTS OF ISPM
This section explains the design requirements identi-
fied in section 4. This section is divided into five sub-
sections, each for one group of design requirements.
The five subsections are further divided into thirteen
sub-subsections each for one design aspect of ISPM.
5.1 Contracts
An ISPM will be a marketplace to buy and sell con-
tracts with underlying security events. The contracts
must be specific and all the details regarding the de-
cision criteria, payoff, settlement date, etc. should
be specified before the contract is made available for
trading. The contract price will be an approximate
measure of the probability, mean or median of the un-
derlying events at any time. The ability to use market
prices as forward-looking indicators of security prop-
erties will help in establishing information symmetry
between buyers and sellers (i.e. build a quality sig-
nal), and help security stakeholders to make better and
more informed decisions, by differentiating mediocre
security products from good ones. The contract types
and specifications are explained below.
5.1.1 Contract Types
On the basis of payoff mechanism, the contracts can
broadly be categorized into three types: (i) Binary
contracts, (ii) Index contracts, and (iii) Spread con-
tracts. A comparison of contract types is shown in
Table 2.
Binary Contracts: In binary contracts the payoff
is linked to the occurrence or non-occurrence of
the underlying event. For example, A binary con-
tract that pays $100 to the buyer of the contract
depending on whether the biometric system of
Table 2: Comparison of Contract Types.
Contract
Type
Payoff Mechanism Market
Belief
Binary All or Nothing Probability
Index Proportionate to out-
come
Mean
Spread Double if outcome ex-
ceeds cutoff else noth-
ing
Median
smartphone ’XYZ’ will be found vulnerable to a
spoofing attack by the 31/Mar/2015.If somebody
thinks that the vulnerability does not exist or will
not be found by the contract expiry date, then the
individual will sell the contract. The price of con-
tract will be in the range of $1 to $100, depending
on the beliefs of market participants. For binary
contracts, with a settlement value of $100, the ac-
tual settlement value will be $0 or $100.
Index Contracts: In case of index contract, the
value of the contract is linked to the outcome of
the underlying event. For example, an index con-
tract that pays $0.01 for every data breach dis-
closed between 01/Mar/2015 to 31/Mar/2015, and
if A thinks that 40 or more such incidents will
be disclosed and B thinks that 80 or less such
incidents will be reported, then the market price
will be between $0.39 and $0.81. A will buy for
strictly less than $0.40 (i.e. $0.39), and B will sell
for strictly more than $0.80 (i.e. $0.81).
Spread Contracts: Spread contracts can be used
to bet on whether the outcome related to the un-
derlying event will be above or below a certain
value, i.e. the spread. In spread contracts, the cur-
rent market price can be interpreted as the traders’
expectation of the median outcome of the under-
lying event. Let us consider a contract with the
underlying event being the discovery of a vulner-
ability in a software. Let us say the market maker
is quoting at the price (spread) $50-$55, i.e. the
market maker wants to buy at $50 and sell at $55.
If a trader believes that the price will go up (i.e. a
vulnerability will be discovered) he will buy from
the market maker at $55. If after some time, some
new relevant information is known to the mar-
ket participants and the current quote by market
maker is $60-$65. Then, if the trader who pur-
chased the contract at $55 wants to cash out his
profits, he will sell his contracts at $60 to the mar-
ket maker.
DesignandPerformanceAspectsofInformationSecurityPredictionMarketsforRiskManagement
275
5.1.2 Contract Specifications
An important aspect of contract design is the precise
specification of the contract. The contract specifica-
tions should clearly mention the expiry date, settle-
ment date (settlement date can be different from the
expiry date), payoff, decision criteria and other fac-
tors relevant to the underlying event.
For instance, for a contract which is meant to al-
low betting on discovery of vulnerability ’V’ in the
software ’S’ on or before the date ’D’, the contract
specifications must clearly define the vulnerability
’V’ as in what is included and what is not. Secondly,
the specification should include information on the
source which will be considered for the acceptance
or rejection of the vulnerability. The source can be
a government body regularly reporting such informa-
tion, appearance of the said information in the regular
media, reporting by responsible organization such as
Google through its ’ProjectZero’ (Google, 2014), or a
direct reporting by the vulnerability discoverer to the
market operator.
If the vulnerability can be directly reported to the
market operator then the contract should clearly spec-
ify the testing procedure or the testing body which
will certify the existence of the vulnerability. Further,
a proper policy should be crafted regarding the re-
sponsible reporting of vulnerability to the vendor and
the time period during which the vulnerability infor-
mation will be kept secret. In such a scenario when
the vulnerability ’V’ is directly reported to the market
operator and it is reported on or before the date ’D’
but it needs to be kept secret for sometime, then the
contract will expire on the date mentioned as expiry
date specified on the contract but the settlement of the
contract will take place only after the vulnerabilityhas
been tested and reported to the vendor. It is not nec-
essary to wait for the vulnerability patch. The market
operator may set a fix period within which the ven-
dor is expected to release the patch. However, if the
vendor fails to fix the vulnerability within the given
period, the contract will be settled and vulnerability
information (full or partial) will be made public.
As the information security events and therefore
the information security contracts are different from
the traditional sports or political contracts, it is im-
portant to clearly define the underlying event and the
decision criteria, otherwise an ambiguity in specifica-
tions may lead to disputes and confusion. The con-
cern for ambiguity and confusion is likely to be an-
ticipated by the buyers and sellers. Thus in addition
to ’noise’ during the clearing process, buy/sell prices
will be affected by the concern. The cost of trust
(of outcome decision) will be factored into prices of-
fered by the buyer and seller, thus with low trust, there
might not be any trade.
A closely related real world example is from
TradeSports (TradeSports, 2015). In 2006 Trade-
Sports had a contract with a payoff value of 100
if North Korea would launch a test missile and the
missile leaves North Korean air space on or before
11:59:59pm ET on 31st July 2006 otherwise the will
pay 0 (EOG, 2006). The source to be used for the
settlement of the contract was the U.S. Department of
Defense. In early July 2006, the Government of North
Korean claimed to have conducted such a test and this
was reported by various news sources as well. How-
ever, the U.S. Department of Defense did not confirm
the test leading to the settlement of the contract at 0.
Had the contract mentioned some other source for the
decision the contract could have settled at 100. On the
other hand if the contract had multiple sources of con-
firmation then this could have led to disputes between
the traders and the operator.
5.2 Trading Process
Trading process is divided into two parts, namely
trading mechanism and anonymous trading, and are
explained below.
5.2.1 Trading Mechanisms
The selection of trading mechanism depends upon the
type of contract and its specification. As the informa-
tion security prediction markets are meant to provide
a risk management mechanism, it is extremely impor-
tant that the market participants are allowed to adjust
their bets based on the latest information they may
have about the underlying event.
(Pennock, 2004) identified four types of trad-
ing mechanisms, (i) Continuous Double Auction; (ii)
Continuous Double Auction with Market Maker; (iii)
Pari-mutual Market; and (iv) Market Scoring Rule.
However, Pari-mutual market mechanism does not
allow continuous incorporation of information and
therefore it is not useful for the information secu-
rity prediction market. (Pennock, 2004) proposed a
new trading mechanism called Dynamic Pari-mutual
Market, and this mechanism provides the continuous
incorporation of information. These trading mecha-
nisms are compared on the basis of four criteria: (i)
Continuous incorporation of information; (ii) Liquid-
ity guarantee; (iii) Ability to cash out anytime during
the market trading hours; and (iv) Bounded risk to
market operator.
Continuous Double Auction (CDA): CDA is a
widely used mechanism in the financial markets.
In this model, limit order book is used to match
SECRYPT2015-InternationalConferenceonSecurityandCryptography
276
buy orders with the sell orders. The orders are
matched based on the price and time priority.
Apart from the limit orders, any trader can buy
at the best ask price and sell at the best bid price.
Continuous Double Auction with Market Maker
(CDAwMM): CDAwMM is a bookie mechanism
which is used in sports betting. This mecha-
nism is like CDAwith an automated marketmaker
which guarantees market liquidity by transferring
the risk to the market operator. The mechanism
allows continuous incorporation of information.
Dynamic Pari-Mutuel Market (DPM): DPM was
proposed by (Pennock, 2004). In DPM mecha-
nism, traders are allowedto purchase shares at any
time and on any outcome. This mechanism pro-
vides an automated market maker which provides
infinite liquidity to the buyers. The trading prices
are set by the market maker on the basis of a pre-
determined pricing formula. The pricing of con-
tracts allows continuous incorporationof informa-
tion. However, the market maker does not buys
the shares from traders and the CDA mechanism
is used to allow selling of contracts by traders.
Market Scoring Rule (MSR): MSR was proposed
by (Hanson, 2003) for trading in prediction mar-
kets . He suggested that a scoring rule can be used
to allow betting on the entire probability distribu-
tion over many variables and a scoring rule will
reward the traders for incremental improvements
in the outcome. All the traders are allowed to see
the current probability distribution and if a trader
thinks that the current probability is incorrect then
the trader can readjust the probability. In this case,
each new predictor is paid off for improving the
prediction probability and traders lose money if
their prediction is worse. The final pay off de-
pends on the closeness of trader’s predictions to
the actual outcome.
Table 3 presents a comparison of trading mechanisms.
Table 3: Comparison of Trading Mechanisms.
CDA CDA
wMM
DPM MSR
Continuous
Information
Incorporation
Yes Yes Yes Yes
Liquidity Guar-
antee
No Yes Yes Yes
Anytime Cash
Out in Market
Hours
Yes Yes Yes Yes
Bounded Risk to
Market Operator
Yes No Yes Yes
5.2.2 Anonymous Trading
Anonymous trading implies that the identification in-
formation about traders is not revealed to market par-
ticipants. In anonymous trading system, traders with
private information are more likely to participate, as
it would be extremely difficult for other traders to dis-
tinguish between the orders and trades coming from
an informed and an uninformed trader. Furthermore,
threats of retribution by the employer of the trader can
deter the trader participation. On the other hand, secu-
rity researchers working in the field may face certain
ethical concernsor peer pressure against the participa-
tion in the market. In such cases, anonymous trading
will let the insiders with private information to trade
in the market. This will improve the information ag-
gregation and the overall performance (efficiency) of
the market.
5.3 Participants and Incentives
Market participants and participation incentives are
two keydesign aspects of informationsecurity predic-
tion markets. These two factors are explained below.
5.3.1 Participant’s Motivations
ISPMs are expected to attract at least six types of
participants: (i) Product Users; (ii) Cyber-Insurance
Providers; (iii) Investors; (iv) Product Vendors;
(v) Product Vendor Competitors; (vi) Security Re-
searchers. Each of these users may have unique or
similar motivations to participate in the market. For
instance for a contract such as : Vulnerability ’V’ will
be discovered in Product ’P’ on or before 11:59:59
PM(GMT) 31/Mar/2015. The motivation for each of
the participant type to trade in the above contract is
shown in Table 4.
5.3.2 Incentive Structure
An appropriate incentive structure is required to mo-
tivate traders to participate in the market and truth-
fully reveal the information relevant to the underly-
ing the event. The incentives for participation in a
prediction market are either monetary incentives or
non-monetary incentives. However, as the goal of
the ISPM is to provide a risk management mecha-
nism for nancial impact associated with the under-
lying information security events, the incentive struc-
ture in this case can only be monetary. The incentive
structure will vary based on the payoff mechanism of
contracts. Different contracts, such as binary, index,
spread, bonds, insurance-linked, etc. will have differ-
DesignandPerformanceAspectsofInformationSecurityPredictionMarketsforRiskManagement
277
ent incentive structure and some contracts may be of
special interest to specific type of participants.
Table 4: Participants and Incentives.
Participants Incentives
Product Users To hedge the productrisk
Cyber Insurance
Providers
To underwrite their cus-
tomer’s cyber-risks
Product Vendors To prove the security of
their software, to sig-
nal its quality with re-
spect to other vendors
products,to use it as em-
ployee stock options to
incentivize developers to
develop secure products
Product Vendor
Competitors
To prove that the product
is not as secure as their
own product
Security Researchers To earn profits from the
market
5.4 Clearing House
The clearing house is a body which clears and set-
tles all the trades in the market. In an ISPM, clearing
house will play at least the following four roles.
5.4.1 Contract Settlement
Due to very technical nature of information security
contracts it will not be easy and straight forward to
decide on the technicalities. For example, in case of a
contract associated with privacy breach at a bank, it is
important to specify and define as in what constitutes
the privacy breach and what is not covered in the con-
tract specification. Thus, the clearinghouse will act as
the final decision body with regards to the settlement
of the contracts. Further, for the contracts related to
vulnerability discovery and when the vulnerability is
directly reported to the market (i.e. to the clearing
house) they have the responsibility to test and accept
or reject the vulnerability. This will lead to settlement
of the contract and final pay outs.
5.4.2 Counterparty Risk Management
The clearing house will act as a guarantee between the
two counterparties involved in the trade. This implies
that the clearing house acts as a buyer for the seller of
the contract and as a seller for the buyer of the con-
tract. In case of zero net supply contracts, the clearing
house has no market risk. However, it is exposed to
the risk associated with failure of any of the counter-
party in the market. Thus, the clearing house needs to
have sufficient capital to cover the risk.
5.4.3 Trusted Third Party
The clearing house will act as a trusted third party for
various activities including but not limited to verifi-
cation of vulnerabilities directly reported to it, keep-
ing the vulnerability information secret for a specified
period, responsibly reporting the vulnerability to the
vendor, etc.
5.4.4 Intellectual Property Management
The ISPMs are likely to have a range of contracts
with various types of underlying events or informa-
tion. Thus, for contracts involving some kind of in-
tellectual property, such as development of an exploit
against the specified discovery will need to be dealt in
an ethical and legal way.
5.4.5 Know Your Trader
Due to the security concerns associated with the infor-
mation related to the underlying events and financial
aspect of trading (incentives), it is utmost important
for the market (clearing house or a separate entity) to
verify the credentials of the market participants. This
is like the know-your-customerpolicy followed in the
banking and financial industry.
Currently, it is extremely difficult to track down
the origin of specific exploit trading in a black mar-
ket. Black-hat hackers operate covertly to succeed in
the game of attack and defense. This intense secrecy
makes it difficult to track and prosecute the exploiter.
So, to prevent the leakage of vulnerability and/or ex-
ploit information from the ISPM to a black-market, it
is must to have a participant’s verification policy.
Know-Your-Trader policy will also help in pre-
venting insider trading. In the financial services sec-
tor, insider trading is defined as the trading of stocks
and other securities of a publicly listed company
by individuals with access to non-public information
about the company. In the information security do-
main, an insider trading can take at least two forms:
(i) A developer working for a company deliberately
leaves some vulnerability in the code, so that it can
be later sold in the ISPM, (ii) A software tester while
working for his/her employer may find some bugs in
an application, however instead of reporting it inter-
nally the individual may decide to sell the information
in the ISPM.
Thus, the traders participating in information se-
curity prediction markets will have to provide some
SECRYPT2015-InternationalConferenceonSecurityandCryptography
278
(under the prevailing laws and regulations) informa-
tion about their finances (to avoid credit default, il-
legal financing, etc.) and personal information. The
personal information is required to fix the account-
ability and responsibility on the trader. If the private
information held with a trader is used for some illegal
activity then the trader can be investigated and held
liable for the same. However, none of this informa-
tion should be made available to other traders in the
market.
5.5 Market Management
The two key aspects of management of information
security prediction markets are regulations and legal
permissions, which are explained below.
5.5.1 Regulated Market
Economic markets can broadly be classified into two
forms, i.e. Free Markets and Regulated Markets. In
a free market economy, the forces of supply and de-
mand are not controlled by a government or authority.
In contrast to a free market, in a regulated or con-
trolled market, government intervenes in supply and
demand through non-market methods such as laws to
control the permissions to participate in the market,
setting of prices, type of products or services, taxa-
tion, etc.
Markets for vulnerabilities will have several issues
such as intellectual property, rights management, non-
disclosure agreements, privacy issues, country spe-
cific laws, industry specific laws, etc. to be dealt in
a fair manner. Thus, the information security predic-
tion market will in most cases be regulated.
5.5.2 Legal Permissions
Currently, there are two main legal issues associ-
ated with information security prediction markets: (i)
compliance with the financial market and gambling
regulations; and (ii) regulations concerning the infor-
mation (cyber-) security.
6 PERFORMANCE MEASURES
FOR ISPM
This section presents the six performance measures
required for effective and efficient functioning of
ISPM, and are explained in the following subsections.
6.1 Information Elicitation
In an ISPM the price of a contract will be informative
only when it is strongly correlated to actual proba-
bility of occurrence of the underlying event. We use
the term ’informative price’ to indicate that there is
a strong - and publicly known- correlation between
price and the aggregated belief of the stakeholders.
Informative prices are extremely vital for the risk
managers to allocate resources (deploy security con-
trols) in an efficient and effective manner.
In financial terms there are two types of values,
i.e. market and intrinsic value, associated with an in-
formation security contract. The market value of a
contract is the current price at which trades can be
executed. The intrinsic value is the expected present
value of the contract with all the available informa-
tion, and accounting for the benefits and costs asso-
ciated with the contract. Since all the information is
not known to all the traders and they differ in their
analysis of information, they have different estimates
about ’true’ probability of occurrence of the underly-
ing event. Therefore, intrinsic values are not perfect
foresight indicators.
Informed traders will estimate the probability of
the underlying event based on the private information
as well as the information available in the public do-
mains. If the intrinsic value estimated by informed
traders is different from the current market price, then
they will buy the undervalued contracts and sell the
overvalued contracts. This difference between the in-
trinsic value and market price is due to noise. There-
fore, the trading by informed traders leads to market
prices closer to the intrinsic value calculated by them.
Let us assume that there are N traders trading a
contract and each of them has a different estimate
of occurrence of the underlying event. Let Pi be the
probability estimate of the i’th trader. Further, let us
say P is an unbiased estimate of T the true probability
of the underlying event. In this case, the forecasted
probability can be expressed as:
P
i
= T + E
i
(1)
where Ei is the error in the probability estimate of i’th
trader. As the probability estimates are unbiased, the
expected forecast error is zero. However, in absolute
terms the individual probability estimate errors might
be quite high.
Let us assume that for each trader the desired po-
sition in the contract Ci is proportional to the differ-
ence between his probability estimate and the current
market price (which indicates probability estimate of
occurrence of underlying event), expressed as:
C
i
= µ(P
i
M) (2)
DesignandPerformanceAspectsofInformationSecurityPredictionMarketsforRiskManagement
279
where µ is some constant of proportionality, and M is
the current market price. Thus, the i’th trader would
like to buy the contract if his probability estimate is
higher than the current market price. On the other
hand, if the trader’s estimate of probability is less than
the current market price then the trader will (short)
sell the contract to profit from the current higher mar-
ket price (probability).
Further, let us assume that the contract is in ’zero
net supply’. Derivatives like futures and options are
examples of zero net supply financial products. Zero
net supply means that for every winner there is a loser
in the market. Thus, if we sum up the market value
of all the position holders in the market, then we get
an exact zero. This implies that as a whole the market
is not exposed to any market risk, however this is not
true for counterparties individually. Zero net supply
is important because it ensures that there is no upper
limit to the scale of the market.
The current market price for zero net supply con-
tracts can be calculated by summing all the desired
positions equal to the net supply and computing the
resulting equation for M (Harris, 2002):
N
i=1
C
i
=
N
i=1
µ(P
i
M) = µ
N
i=1
P
i
NµM = 0 (3)
The market price M is represented as
M =
1
N
N
i=1
P
i
(4)
is an average of the individual probability estimates of
underlying events. Substituting Equation 1 into Equa-
tion 4 gives:
M = T + E
mp
(5)
where Emp is the market price forecast error repre-
sented as follows:
E
mp
=
1
N
N
i=1
E
i
(6)
If the forecast errors of individual traders is indepen-
dent of each other, then the law of large numbers will
come into play and with an increase in the number of
traders, the market price forecast error Emp will ap-
proach zero. Also, if the number of traders is small
and if the individual trader errors are not identical
then the average market price forecast error will be
less than the average market price forecast error of in-
dividual traders. Therefore, the information security
prediction markets will be most informative if the in-
formed traders independently collect the information.
6.2 Transparency
Transparency is an important characteristics. Trans-
parency deals with the information about the trading
process such as prices, book size, etc. which is made
available to the traders. The market transparency will
ensure that the clearing house and market regulators
are aware of the positions of individual traders. Thus,
if a market participant becomes too exposed to mar-
ket risk or is accumulating big positions which could
affect the overall market, then the authorities can ini-
tiate the necessary risk management and legal actions.
Transparency in an ISPM can be divided into pre-
trade transparency and post-trade transparency. Pre-
trade transparency is about the information on bids,
offers, book size, order depth and other such infor-
mation which is useful before a trade has been made.
Post-trade transparency is about providing informa-
tion related to executed trades, such as time of trade,
price at which it was executed, size of the trade,
etc. The identity of traders should not be revealed
to other traders, neither in pre-trade information nor
in post-trade information. Further, Information re-
lated to vulnerability and/or exploits (if it has been
directly reported to the market operator) cannot be
fully disclosed to market participants until a patch has
been released or the vendor was given adequate (as
mentioned in the trading contract description) time to
fix the same. However, historical trading prices, or-
ders and other trade related and settlement informa-
tion should be made available to the market partici-
pants. This information has twofold benefits. First,
it lets the traders analyze the historical information
and secondly the information can be used by cyber-
insurance and information security rating companies
for the pricing of insurance contracts and rating of
product vendors respectively. Higher transparency in
the market is expected to lead to transparent pricing
of contracts.
Further, information security researchers cur-
rently face the problem of lack of data to validate
various models of information security investments in
security controls, estimating security strength of a se-
curity control, reputation of security product vendors,
and so on. The data released by information secu-
rity prediction markets will be highly useful in such
scenarios where researchers and practitioners face the
problem of lack of data. The data can further be use-
ful in devising successful risk mitigation strategy, de-
velopment of secure software, among various other
benefits.
6.3 Efficiency
In an efficient market, market prices reflect all the in-
formation that can be acquired by traders and prof-
itably acted upon (Fama, 1970). However, some in-
formation may be too expensive to acquire or of lit-
SECRYPT2015-InternationalConferenceonSecurityandCryptography
280
tle value if it cannot be used to profit from the trade.
Therefore, market prices can never reflect all the in-
formation which informed traders can collect and act
on. The market efficiency can be categorized into
three types (Fama, 1970):
1. Weak Efficient Market: In the weak-form of ef-
ficient markets, the market prices reflect all the
past information and no one can make profit by
knowing the past information only. In this case,
prices simply follow a random walk. Therefore,
this form of market efficiency is not useful for in-
formation security prediction market.
2. Semi-strong Efficient Market: In the semi-strong
form of efficient markets, the market prices reflect
all the information available in the public domain.
In this case, no one can predict the future market
price by using only the information in public do-
main. In semi-strong form of markets, informed
traders can profit by having access to some pri-
vate information. In an ISPM security researchers
may have private information about certain vul-
nerability in a software and they can profit from
it. Also, if a contract is listed for discovery of
a vulnerability in a particular software then secu-
rity researchers can use their domain knowledge
to discover the said vulnerability before the ex-
piry date, thus making profit from trading in the
market.
3. Strong-Form Efficient Market: In the strong form
of efficient markets, the market prices reflect all
the information available in the public domain
as well as the private information as soon as it
is known. In such a scenario, informed traders
have no advantage over uninformed traders and
therefore they can never make profit in the mar-
ket. Therefore, this type of market efficiency is
not useful for ISPMs.
6.4 Transaction Cost
Transaction cost will include all the cost associated
with trading in the ISPM. These costs can be divided
into following three categories (Harris, 2002):
Explicit Transaction Costs: This includes costs
like brokerage paid, exchange fees, taxes paid.
This also includes cost of acquiring relevant in-
formation such as information related to a vulner-
ability, software product, etc. Further, it includes
the cost associated with time spent and resources
used by a security researcher to discover vulner-
ability in a software. The security researcher can
use this private information about the existence of
vulnerability in the software to trade relevant con-
tract(s) listed on the information security predic-
tion market.
Implicit Transaction Costs: This type of trading
cost arises when traders have an impact upon the
market prices. For instance, if a trader buys at the
ask price and sells at the bid price then he ends up
paying the bid-ask spread price. Similarly, when
traders push up the price while executing large
buy orders, and push down the prices when exe-
cuting a large sell order, the impact of their trading
on the prices constitutes transaction cost.
Missed Trade Opportunity Costs: This is the cost
associated with failure to get orders executed or
if the orders are partially filled or if the orders
are not filled in a timely manner. In other words,
missed trade opportunity cost is the difference be-
tween getting the trade at the desired price, and
the first next opportunity. So, it is the cost associ-
ated with delay in transaction.
In the financial industry, the simplest and commonly
used method for calculation of transaction costs is
’Quoted Spreads’ (Teschner, 2012). This can be cal-
culated using the trade and order book data. Let us
say Bidc,t is the bid price for a contract ’c’ at time ’t’
and Askc,t is the corresponding ask price for the con-
tract ’c’ at time ’t’. The mid quote of Mid price of the
contract c is denoted as Midc,t. The quoted spread can
thus be calculated as:
QuotedSpread
c,t
=
(Ask
c,t
Bid
c,t
)
2Mid
c,t
(7)
The spread paid when executing a market order
against a limit order is termed as effective spread. Let
us say Pricec,t is the execution price then the effective
spread can be calculated as:
E f f ectiveSpread
c,t
= S
c,t
(Price
c,t
Mid
c,t
)
Mid
c,t
(8)
Sc,t denotes the trade side, +1 for a buy order and -1
for a sell order.
The realized spread denotes the revenues of liq-
uidity supplier (Bessembinder and Kaufman, 1997).
(Glosten and Milgrom, 1985) model highlights that
if the risk of trading against asymmetrically informed
traders is high then the spread is wide to compensate
the informed traders for their losses. After ’n’ minutes
of trade execution, the realized spread is calculated as:
RealizedSpread
c,t
= S
c,t
(Price
c,t
Mid
c,t+n
)
Mid
c,t
(9)
One of the recently proposed spread estimator method
which can be used for ISPMs is (Corwin and Schultz,
2012). As all the orderbook data may not be available
DesignandPerformanceAspectsofInformationSecurityPredictionMarketsforRiskManagement
281
or the data availability will be limited in the ISPM, the
transaction price method proposed by (Corwin and
Schultz, 2012) will be useful. They derived an esti-
mator for the bid-ask spread based on the daily high
and low prices. Daily high prices Hp is almost always
the execution price of buy order and daily low price Lp
is most likely the execution price of a sell order. The
price ratio of high-to-low price is due to the volatil-
ity which proportionately increases with the length of
trading intervals. Hence, they proposed an estimate of
a contract’s bid-ask spread as a function of the high-
to-low price ratio for a single two day period. The
high-to-low ratio for two consecutive trading days is
(Hp, p+1, Lp, p+1). They defined the spread estimator as:
SpreadEstimate =
2(e
α
1)
1+ e
α
(10)
where
α =
p
(2β)
p
β
32
2
r
γ
32
2
(11)
β = (ln
H
p
L
p
)
2
+ (ln
H
p+1
L
p+1
)
2
(12)
γ = (ln
H
p, p+1
L
p, p+1
)
2
(13)
6.5 Liquidity
Liquidity can be defined as the ability to execute large
size orders quickly, at low cost and at any time during
the market hours (Harris, 2002). Liquidity is an im-
portant market characteristic for market stakeholders.
High liquidity allows traders to execute their trades
in an efficient and cost-effective manner. Market op-
erators like a liquid market because it attracts many
traders to the market, and improves liquidity and in-
formation efficiency in the market. Market regulators
like liquid markets as the liquid markets are often less
volatile than the illiquid markets.
The (Amihud, 2002) illiquidity measure can be
used for (il)liquidity measurement in the ISPM. The
method uses the daily ratio of absolute returns for a
specific contract at a given time, in relation to the trad-
ing volume. Let us say that |Rct| represents the return
for contract ’c’ in the time period ’t’, and Volct rep-
resents the trading volume for the contract ’c’ in the
time period ’t’, and TDi denotes the number of trading
days in that period. The illiquidity for the contract ’c’
represented as ILc is expressed as:
IL
c
=
1
T
Di
T
Di
t=1
|R
ct
|
Vol
ct
(14)
Let us say, there is a futures contract listed on ISPM
which pays $1 if a vulnerability is discovered in the
biometric authentication system of mobile phone ’M’
on or before X’ date and pays nothing otherwise. If
all the traders who do not have any private informa-
tion and believe that there is 40 percent probability of
discovery of vulnerability in the said system and on or
before the expiry date, then the price of the contract
will be 40 cents.
On the other hand an information security re-
searcher who has some knowledge about the system
believesthat he can discovera vulnerability before the
expiry date, Thus for him the probability of vulnera-
bility discovery is higher than 40 percent. Thus the se-
curity researcher believes that the probability of vul-
nerability discovery stands at 100%, denoted by λ’.
In this case the security researcher is well informed
and no one else has this information. So, when he
starts buying contracts starting from 40 cents, he is
pushing up the price.
Let us say, that average buy price for the security
researcher is represented as:
P
AVB
= 0.4+
T
c
L
(15)
where PAVB is the average buy price, Tc denotes the
total number of contracts purchased, and L is a pa-
rameter that characterizes the market liquidity. If the
market is very liquid then the security researcher can
buy a large number of contracts without significantly
affecting the market price.
The number of contracts which the security re-
searcher should buy to maximize his profits depends
upon λand L. As the contract pays $ 1 with probabil-
ity λ, the expected value of owning the contract is λ.
Therefore, the expected value of security researcher’s
holding is λTc. The total cost incurred on acquisi-
tion of this position stands at PAVBTc. So the expected
profit for the security researcher can be represented
as:
λT
c
(P
AVB
T
c
) = λT
c
(0.4T
c
+
T
c
L
)T
c
(16)
The expression shows that the security researcher
with good information can make more money in a liq-
uid market.
6.6 Manipulation Resistance
Manipulation in information security prediction mar-
kets can take at least two forms. Firstly, when a trader
tries to control the rate of information revelation and
decides to temporary trade against the information. A
trader who is confident that no one else has the same
information may prefer to do several small size trades
instead of one big trade. Also, to avoid signaling to
other traders, the trader needs to avoid pattern, such as
SECRYPT2015-InternationalConferenceonSecurityandCryptography
282
same order size in his small size trades (Chakraborty
and Yilmaz, 2004). This form of manipulation (hiding
true information) is relatively less harmful. However,
if the trader fears that other traders may also acquire
the same information then the trader may actually do
a big trade thereby revealing (signaling) the informa-
tion to other traders.
A second type of manipulation of prices is pos-
sible when traders trade in opposite of the informa-
tion they have and later the trades are not reversed.
Traders sometimes do this to create confusion in the
market and let the contracts trade at or around a par-
ticular price. This type of manipulation can be tack-
led by limiting the trade limit of traders, so that if
they trade against the information they will lose on
the trades. Also, if they do not trade quickly on the in-
formation they have then other may acquire the same
information and trade accordingly thereby deepening
the losses for those who did not trade or traded in
the opposite direction. Researchers have studied var-
ious models of financial market microstructure with
consideration of various types of noise traders such
as those who trade randomly, traders who have ma-
nipulate the closing price for higher futures market
settlement, traders who need immediate liquidity, and
traders who have quadratic preferences over the cur-
rent market price (Kumar and Seppi, 1992; Hillion
and Suominen, 2004; Hanson et al., 2006).
These studies have proved that manipulators are
just another type of noise traders and in effect they
improve the price accuracy. A trader seeking to ma-
nipulate the prices has hidden bias towards the direc-
tion and extent to which he intends to manipulate the
price. Other traders participating in the market are ex-
pected to have average bias. When the manipulator’s
bias and average trader bias is exactly equal then the
markets perform as if there is no price manipulation.
On the other hand, if the manipulator bias is lower or
higher than the average bias then this will be reflected
in market prices. However, speculators competition
leads to correction in average price and the price ac-
curacy is not affected with manipulative noise trading.
The studies based on market data confirms the theory
that average price accuracy is not affected by manip-
ulators. Though there has been one case of successful
manipulation (Hansen et al., 2004) but others have re-
ported that manipulators have not been successful in
decreasing the price accuracy, in the field (Camerer,
1998), historically (Rhode and Strumpf, 2004), and
in the laboratory (Hanson et al., 2006).
The evidence that noise trading generally leads to
increase in market accuracy suggests that there is little
to worry about price manipulation attempts in the in-
formation security prediction markets. As long as the
traders with correct information and rational behavior
outnumber the manipulators in terms of trading size,
the net effect will lead to increase in average price
accuracy. In other words, the impact of price manip-
ulation in information security prediction market will
depend upon the financial muscles of the manipulator.
7 CONCLUSIONS AND FUTURE
WORK
In this article, we identified a set of design require-
ments for the ISPM. We explained three types of con-
tracts: binary, index, and spread. Also, we explained
the importance of clear and unambiguous specifica-
tion of contracts. We explained four different types
of trading mechanisms, CDA, CDAwMM, DPM, and
MSR. The trading mechanisms are compared on the
four criteria: continuous incorporation of informa-
tion, liquidity guarantee, bounded risk to market op-
erator, and cash out option during the market hours.
The specific features of the trading mechanisms can
be used to design contracts to achieve the specific
objectives. We explained the significance of clear-
ing house in settlement of trading contracts, coun-
terparty risk management, role of trusted third party,
and as a custodian of confidential information and in-
tellectual property. Also, we identified several types
of traders who are expected to participate in ISPM.
Also, trader’s motivation and incentivesfor the partic-
ipation is explained. The importance of market reg-
ulation and specific legal permissions is briefly dis-
cussed. Furthermore, we identified six performance
measures for ISPM: Information elicitation, Trans-
parency, Efficiency, Transaction Cost, Liquidity, and
Manipulation resistance. We explained these individ-
ual factors in the light of information security specific
objectives of the prediction market.
In future, we intend to design, demonstrate and
evaluate some financial contracts to address a partic-
ular type of security risks. Then, we plan to demon-
strate a complete design and model of risk hedging in
an information security prediction market.
REFERENCES
Amihud, Y. (2002). Illiquidity and stock returns: cross-
section and time-series effects. Journal of Financial
Markets, 5(1):31–56.
Berg, J. E. and Rietz, T. A. (2003). Prediction markets as
decision support systems. Information Systems Fron-
tiers, 5(1):79–93.
DesignandPerformanceAspectsofInformationSecurityPredictionMarketsforRiskManagement
283
Bessembinder, H. and Kaufman, H. M. (1997). A cross-
exchange comparison of execution costs and informa-
tion flow for NYSE-listed stocks. Journal of Financial
Economics, 46(3):293–319.
Camerer, C. F. (1998). Can asset markets be manipulated?
a eld experiment with racetrack betting. Journal of
Political Economy, 106:457–482.
Chakraborty, A. and Yilmaz, B. (2004). Manipulation in
market order models. Journal of Financial Markets,
7(2):187–206.
Corwin, S. A. and Schultz, P. (2012). A simple way to es-
timate bidask spreads from daily high and low prices.
Journal of Finance, 67(2):719–760.
EOG (2006). Tradesports’ bad call.
http://forums.eog.com/showthread.php/39353-
Tradesports-Bad-Call.
Fama, E. F. (1970). Efficient capital markets: A review of
theory and empirical work. The Journal of Finance,
25(2):383–417.
Fidler, M. (2014). Anarchy of Regulation: Controlling the
Global Trade in Zero-Day Vulnerabilities. PhD thesis,
Stanford University.
Glosten, L. R. and Milgrom, P. R. (1985). Bid, ask and
transaction prices in a specialist market with heteroge-
neously informed traders. Journal of Financial Eco-
nomics, 14(1):71–100.
Google (2014). Projectzero.
http://googleprojectzero.blogspot.be/.
Hansen, J., Schmidty, C., and Strobelz, M. (2004). Manip-
ulation in political stock markets - preconditions and
evidence. Applied Economics Letters, pages 459–463.
Hanson, R. (2003). Combinatorial information market de-
sign.
Hanson, R. (2006). Designing real terrorism futures. Public
Choice, 128(1-2):257–274.
Hanson, R., Oprea, R., and Porter, D. (2006). Information
aggregation and manipulation in an experimental mar-
ket. Journal of Economic Behavior & Organization,
60(4):449–459.
Harris, L. (2002). Trading and exchanges: Market mi-
crostructure for practitioners. Oxford University
Press.
Hillion, P. and Suominen, M. (2004). The manipula-
tion of closing prices. Journal of Financial Markets,
7(4):351–375.
Johannesson, P. and Perjons, E. (2014). An Introduction to
Design Science. Springer International Publishing, 1
edition. ISBN: 978-3-319-10631-1.
Kumar, P. and Seppi, D. J. (1992). Futures manipula-
tion with cash settlement. The Journal of Finance,
47(4):1485–1502.
Luckner, S. (2008). Prediction markets: Fundamentals,
key design elements, and applications. The 21st Bled
eConference, eCollaboration: Overcoming Bound-
aries Through Multi-Channel Interaction.
Pandey, P. and Snekkenes, E. (2014a). Applicability of pre-
diction markets in information security risk manage-
ment. In Database and Expert Systems Applications
(DEXA), 2014 25th International Workshop on, pages
296–300.
Pandey, P. and Snekkenes, E. (2014b). Using prediction
markets to hedge information security risks. In Mauw,
S. and Jensen, C., editors, Security and Trust Man-
agement, volume 8743 of Lecture Notes in Computer
Science, pages 129–145. Springer International Pub-
lishing.
Pennock, D. M. (2004). A dynamic pari-mutuel market for
hedging, wagering, and information aggregation. In In
Proc. of the 5th ACM Conf. on Electronic Commerce,
pages 170–179.
Plott, C. R. and Chen, K.-Y. (2002). Information Aggre-
gation Mechanisms: Concept, Design and Implemen-
tation for a Sales Forecasting Problem. W.P. 1131,
California Institute of Technology.
Rhode, P. W. and Strumpf, K. S. (2004). Historical presi-
dential betting markets. The Journal of Economic Per-
spectives, 18(2):127–141.
Spann, M. (2002). Virtuelle B¨orsen Als Instrument Zur
Marktforschung. Deutscher Universit¨ats-Verlag.
Sripawatakul, P. and Sutivong, D. (2010). Decision frame-
work for constructing prediction markets. In The 2nd
IEEE Int. Conf. on Information Management and En-
gineering (ICIME), 2010.
Teschner, F. (2012). Forecasting Economic Indices De-
sign, Performance, and Learning in Prediction Mar-
kets. PhD thesis, Karlsruher Institut fr Technologie.
TradeSports (2015). Tradesports.
https://www.tradesports.com/.
WEF (2014). Global risks 2014. Insight Report 9th Edn.,
No.: 090114, World Eco. Forum, Geneva.
Weinhardt, C. and Gimpel, H. (2007). Market engineering:
An interdisciplinary research challenge. In Jennings,
N., Kersten, G., Ockenfels, A., and Weinhardt, C., ed-
itors, Negotiation and Market Engineering, number
06461. IBFI, Germany.
SECRYPT2015-InternationalConferenceonSecurityandCryptography
284