ریسک معامله در تجارت الکترونیک

Electronic commerce business models can add value by elimination of control and risk-bearing borne by channel intermediaries. But such markets may be less robust encouraging risky behavior. We explore a model of transaction risk in the migration from broker-mediated to electronic markets. The research finds that: (1) the risk of falsely accepting ‘bad’ orders is the critical risk measure for optimal control choice; (2) control policies and operational policies must be established simultaneously; (3) optimal control choice is strongly influenced by risk preferences; (4) at the margin, the decrease in risk from an additional increment of control is proportional to the market's transaction volume; and (5) the discretionary budget sets the upper limit to control.

This paper provides a model for examining the importance of properly controlling transaction risk in electronic commerce. The paper addresses the question “what level of managerial control is required to reliably assure that payment and services will be rendered, goods delivered, and quality will be adequate.” Transaction risk results when markets fail to provide one or more of these when processing a transaction. Controls are subsystems that limit the frequency or magnitude of damage from failed transactions. This paper assumes that control is applied at the order level and draws conclusions about the level of risk and control required to provide a viable market. Control of transaction risk in electronic commerce has become important because electronic commerce often adds value over its physical counterparts by allowing high transaction volumes at significant operating economies, but at the cost of less flexibility, robustness and control over risk, due to the loss of human channel intermediaries (brokers) in physical markets, usually relegating them to digital surrogates, or just transmission bandwidth. The problem has been highlighted in the recent U.S. Federal Bureau of Investigation inquiry into whether rings of shill bidders on eBay have committed fraud by bidding up the prices of one another's online auction offerings [3]. The problem is considered widespread in Web-based auctions. The control of transaction risk is of increasing concern as electronic markets become pervasive. Electronic markets discard many broker-based controls inherent in traditional markets. The brokerage function has been especially important in high-volume markets such as stock exchanges. The broker's agency1 relationship with the trader allows securities markets to operate at high capacity with high reliability and low risk. In these markets, monitoring of individual transactions is transferred to many individual brokers, leaving the market to concentrate on matching orders. In U.S. securities markets, controls are highly developed with several stages of indemnification. Individuals do not trade directly in these markets, but must trade through brokers who guarantee to the market that the individual can deliver goods or services, and that settlement can be made [2]. The broker may engage a clearing agent on the market floor to assure the broker that buy and sell orders will be matched. The clearing agent is indemnified by a clearinghouse, which assures his or her performance. Counterparts to the brokerage function appear in markets for many non-financial goods and services. Food, for example, is sold by grocery stores that are responsible for purchase, delivery, and assurance of quality and freshness. Liquidity and matching demand with supply are particularly important with perishables. Building contractors procure subcontractors who provide specialized labor for construction as well as assurances that a construction project will be completed properly, on-time and within budget. To assure that brokers can effectively control transaction risk, securities markets set minimum earnings and capital requirements for firms that list their securities on their exchanges, and monitor these brokers extensively. Exchanges that have set slack listing requirements—in recent history the Denver Exchange and the Vancouver Exchange are examples—have seen their reputations erode, and have watched their traders move elsewhere to conduct business. Markets for goods and services over the past decade have invested significantly in automation of various market components through electronic data interchange and other technologies.2 Without automation, markets are constrained to operate at the speed of their human facilitators—frequently too slow for complex or high-volume market services. In order to speed up transaction processing, traditional markets may be stripped of all but market matching functions, and other functions dispersed to brokers, clearing houses and similar operations. Electronic markets have naturally focused on the particular services that are well suited to technology—matching and information dissemination. Much of design has been ad hoc-based more on what technology can do rather than what technology needs to do to provide an efficient, reliable market. There is little research on design to guide appropriate investment [3]. Electronic markets generally assume a direct information link with buyers and sellers—e.g., through VDTs, television sets or kiosks. The structure and implications of this direct linkage have been discussed extensively in the prior literature, e.g., in [1], [2], [4], [5], [6], [7], [11], [14] and [15], and Ref. [12]. Such an architecture eliminates broker control or risk-bearing, and relegates transaction risk control to surrogates built into the electronic market systems. Being software algorithms, surrogate broker systems are not likely to be perfect (or necessarily even close) substitutes for human brokers. Brokers can maintain personal contact with customers, using intuition, experience and judgment to winnow good business from bad. Thus, stakeholders in electronic commerce systems should be interested in the cost of increased transaction risk, and prospects for its control. The subsequent analysis examines risk-bearing and control as markets evolve from broker-mediated to electronic. Section 2 explores the contrasting architectures of broker-mediated and electronic markets. Section 3 presents a formal model of supply and demand for market services. Section 4 discusses risk assessments by market participants. Section 5 computes the optimal control level for a linear model of supply and demand with Gaussian uncertainty. Section 6 draws general conclusions on requirements for control as markets relinquish broker-based controls in their move to automate.

The previous results can now be consolidated into a set of five results that guide control choice as markets migrate from brokered to electronic systems. These results provide insight into electronic commerce systems design, an important consideration as operating cost reductions and attractive novel features make electronic commerce increasingly effective in reaching new customers. (Conclusion 1)β-risk directly increases buyers' and sellers' risks of participating in the market. As a result, β- risk of falsely accepting “bad” orders is the critical risk measure for optimal control choice. Exposures arising from bad orders must also be dealt with after a loss—this is almost always more expensive than rejecting bad orders since the buyer or seller has already suffered the crime. Small value orders for goods and services may be able to support larger quantities of bad orders (and the consequent higher β-risk), but large value items will not be traded in risky markets. To date, electronic markets have had the greatest success in trading low valued goods—e.g., freeware and shareware where marginal cost of the good is close to zero. This is likely a reflection of significantly higher β-risk in existing electronic markets. At certain critical masses of bad orders, two additional effects (outside the scope of this paper) impact buyer–seller participation in the markets: reputation effects, and reduced network externalities as participants leave the market. Poor reputation tends to reinforce itself, and will drive some of the more risk-averse participants away. This lowers the total participation in the market, and thus makes it less liquid forcing longer delays in clearing orders. Thus, it is less attractive for reasons independent of risk. These latter influences reflect the reduction in the market's network externality. The move away from brokered markets toward integrated electronic commerce systems increases β-risk. Electronic commerce, in contrast, favors direct information links which circumvent broker control or risk-bearing. Surrogate risk monitoring systems in electronic markets cannot be perfect (or necessarily even close) substitutes for human brokers. Real brokers can maintain personal contain with customers, using intuition, experience and judgment to winnow good business from bad. This systematic lowering of control in electronic commerce is a trade-off for their faster, cheaper transaction processing. The discriminant choice variable z discriminates between good and bad orders, and through mapping (ξ,z)→(α,β) determines the level of α- and β-risks. If control levels are relatively difficult to change, which is often the case, z alone selects a single point (α,β) on the frontier of efficient decisions. From this perspective, the α-risk of falsely turning good orders away is significantly less important than β-risk in the control of transaction risk. It does not directly influence risk of participating in the market. It does influence long run demand for market services, because it lowers order volume and liquidity making the market less attractive to the potential seller or buyer. It does not lead to poor reputation, nor are there operational problems that impose a cost to the market or its participants. There is always be a certain amount of business that will be rejected, even where most of that business is good business. If this good business is significant, other broker-market communities will accept it and this can lead to a significant loss of revenue. Since the move to electronic commerce is also a move to high-volume, low-margin retailing, the network externality is significant and indirect effects of lower volume can significantly lower demand for market services. The analysis showed that the β-risk of falsely accepting bad orders is most important. β-Risk directly increases buyers and sellers risk of participating in the market. It is likely to be important when large sums of money are at stake, and any gambles in the market are potentially ruinous. Discriminant parameter z should therefore be chosen to minimize β-risk. Note that the ability to efficiently control transaction risk depends on the existence of internal accounting systems that can effectively gather data on, and compute: (i) the “price” traders are willing to pay to avoid α- and β-risks (for use in computing “cost-benefit” risk mapping), (ii) values for θ the actual measure of control, and (iii) the influence of control level on order processing, given by g(θ) and capacity G(θ2). (Conclusion 2)Control policies (which determine control level ξ) and operational policies (which determines the discriminant function's cut-off value z) must be established simultaneously. Setting one policy independently of the other results in unreliable risk assessments and poor control. The control level ξ reflects the effectiveness of resources, at marginal cost κ(ξ), spent limiting risk. Fig. 6a–c shows that ξ multiplies the effect on risk of the discriminant choice variable z on risk ρ. If (α,β)→ρ by “fixed significance level” approaches, then the multiplier is some monotonic function with range (0,1). If (α,β)→ρ by “minimax” approaches, then the multiplier is still in the range (0,1) but has a peak, and will be significantly smaller than the “fixed significance level” multiplier. If (α,β)→ρ by “cost-benefit” approaches, then the multiplier is some monotonic function with range (0,∞) reflecting the inclusion of cost figures which may substantially exceed 1. This approach yields a “price” of risk, in contrast to the other two approaches are probabilistic functions which favor either α and β given a particular situation. In any case, the conclusion that should be drawn from elements 1 and 2 is that control policies (which determine ξ) and operational policies (which determine z) must be established concurrently. Establishing one policy without the other will produce unreliable risk assessments. The effect of each policy is multiplicative, and thus not easily separated from the effect of the other policy. An exception to this conclusion occurs when control level is relatively difficult to change, e.g., when risk arises externally, and is only poorly controlled by managerial decisions. Then risk ρ will depend entirely on the discriminant variable z. (Conclusion 3)Optimal control choice is strongly influenced by the preferences of traders for risk (i.e., risk aversion or risk preference). As new traders enter the market, and as former traders leave, optimal control choice may change substantially. Traders have a demand curve for a particular level of controls, mapped (α,β)→δ(ρ,1−ρ). This may be perceived as an alternative measure of risk aversion. The prior discussion of Fig. 8a–f explored the impact of trader risk aversion. The “fixed significance level” approach led to control choice that was overly sensitive to the risk aversion of traders. The “minimax” approach tended to arbitrarily favor either α- or β-risk in different ranges of the parameters, leading to multiple, conflicting control choices. The “cost-benefit” approach avoided the objections of the other two models by more completely incorporating information about α- or β-risk, but required the collection of information on risk and control preferences of traders. Though this is desirable, trader risk preferences are exogenous to the market, and are likely to be costly to ascertain in the process of setting market policy. The marginal increase in the control ∂ξ̂, i.e., the normalized fraction of market capacity which can be processed at a given control level, is proportional to the marginal decrease in risk ∂ρ at net value maximizing equilibrium, i.e., ∂ρ=−∂ξ̂. This relationship is simple and intuitive but applies to two derived variables. Control ∂ξ̂ depends on the derived ξ̂ which is known only through θ, g(.), and maximum systems order processing capacity G(θ2). Risk ∂ρ depends on discriminant function z, control level ξ̂ and the approach used to map (α,β)→ρ. Fig. 8a–f explored this equilibrium condition. Fig. 8a and b looked at the “fixed significance level” approach. For high values of m, i.e., as transaction risk becomes relatively more importance to traders in the market, the equilibrium levels of control accelerate to extreme heights. This suggests that optimal control choices under this approach would be volatile, and subject to the exogenous whims of traders in the market. While this may indeed be the case, it is difficult to set control policy with a decision model which demands vastly different levels of control expenditure κ(ξ) from day to day as traders preferences change. Fig. 8c and d looked at the “minimax” approach. The approach tends to arbitrarily favor either α- or β-risk in different ranges of the parameters. But prior discussion emphasized the importance of β-risk to traders, while indicating that trading would be much less sensitive to α-risk. Worse, Fig. 8c indicates broad ranges of trader risk preference m in which the model suggests multiple optimal choice of marginal control expenditure κ(ξ). The “minimax” approach, though applicable in certain statistical decision problems, furnishes unrealistic and ambiguous options for control choice. For these reasons, it is not to be recommended as risk mapping model in control choice. Fig. 8e and f looked at the “cost-benefit” approach. This approach provides control choices that are robust (monotonic and smoothly varying) in the parameter m and choice variable z. It avoids the objections to the other two models. In addition, it provides an estimate of loss in (0,∞), rather than a probability in (0,1), and thus bears more information about the choice. This matters, since marginal control expenditure κ(ξ) is likely to be smooth and monotonic increasing, and order processing g(θ) volume at control level θ is likely to be smooth and monotonic decreasing. A smooth monotonic mapping (α,β)→ρ will yield robust choices for optimal control. Optimal control ξ̂ yielding net value maximizing equilibrium ∂ρ=−∂ξ̂ is best realized through a cost-benefit assessment of risk. (Conclusion 4)At the margin, the decrease in risk from an additional increment of control is proportional to the market's transaction volume. Thus, a busy market, with a high volume of trading will benefit more from increased controls; conversely, thinly traded markets will find less need for controls over transaction risk. The marginal decrease in risk ρ is proportional to the marginal increase in the actual measured control θ̂ times the factor g(θ̂)/G(θ2), i.e., the number of orders which can be processed at a given control level θ̂ divided by the market's total ability to process orders at all control levels. The monotonic decreasing function g(θ) reflects the trade-off between tightly controlling risk in a few orders, or loosely controlling risk in an larger number of orders. Optimal control will satisfy the restated equilibrium conditionAt the control margin, the incremental increase in risk from an additional increment of control is proportional to the order processing volume at that level of control. With some reservations, this result can be viewed as assigning a characteristic “packet” of risk to each order processed, and assuming that total risk is approximately additive (which will be true for very small probabilities). This conclusion provides a useful “rule of thumb” for risk assessment. (Conclusion 5)The discretionary budget for spending on controls over transaction risk provides an upper limit to control which may limit optimal control choice. This budget isarginal control expenditure κ(ξ) must come out of a discretionary control budget K(ξ) which is typically fixed for a year. Let K(θ) be the control budget stated in terms of the known measure θ, and recall that the transformation from θ→ξ is ξ=(G(θ))/(G(θ2))⇒θ=G(θ2)G−1(ξ) where G−1 is the inverse function of G. ThenEq. (10a) computes the budget for control at level θ, given G(.) and κ. Eq. (10b) provides entries for the right-hand side of the equilibrium Eq. (2) in terms of G(.) and a priori budget K. The move from a broker-mediated to a fully electronic market holds the potential to dramatically increase processing volume and transaction risk. Thus, the costs and benefits of each of these factors have to be carefully weighed against each other. Increased processing volumes can induce network externalities, which lock traders into the market; but if volume comes at a cost of excessive β-risk, reputation effects may drive traders elsewhere. Since the incremental risk is proportional to order processing volume at a given level of control, network externalities from increased order volume must rise faster than risk if the move to electronic markets is not seen to be excessively risky. Perhaps the current hesitance of individuals to abandon mail order, television and public auctions in exchange for Internet commerce arises from order volume not increasing faster than risk. This may be a problem for some time, since optimal control choice can be highly sensitive to trader risk aversion, at the same time that information about trader risk aversion is difficult to collect. Reputation and excessive β-risk can be limited by control processes and effective selection of orders—both are crucial to market success, and the influence of both is multiplicative. It is advisable that measurement and assessment of risk entail a full cost-benefit model, in order that control expenses not be overly sensitive to market risk tolerance. Where the trading community is not currently using the cost-benefit approach, the research suggests that it should be willing to pay up to its current control expenditures to build or purchase an internal accounting system which can provide the data needed to control transaction risk. In light of this, formal computer algorithms for control, process modeling and accounting should become increasingly important over the next decade. The results of this research may be seen as imperatives for future market design, as well as conditions for choice between brokered and electronic commerce.