Internet pricing: best effort versus quality of service.

Shin, Seungjae ; Cope, Robert F., III ; Cope, Rachelle F. 等

ABSTRACT

This research uses Bertrand methodology to examine the influence of competition between companies that utilize Quality of Service (QoS) pricing strategy versus Best Effort (BE) pricing strategy for Internet Service Providers (ISPs). The Bertrand duopoly price competition model is effective at determining customer's willingness-to-pay and level of internet usage patterns in relation to price paid for service. The model also makes use of a two-part tariff consisting of a fixed rate for Best Effort (BE) service, and a usage-sensitive rate structure for premium QoS. Initial results indicate that an equilibrium market position for each ISP depends on a customer's preference for QoS and the price of BE service. Implementation of this research using a game simulation revealed an analytical framework for iterative, short-term, future QoS Internet pricing strategies.

INTRODUCTION

Since the commercialization of the World Wide Web (WWW), Internet Service Providers (ISPs) have expanded their service simply by increasing their number of subscribers. Traditionally, ISPs offered only one class of service for all types of traffic. They treated traffic indifferently with no discrimination among the types of traffic, no guarantee for timely delivery, and a realistic possibility of traffic loss. This type of Internet service has generally been known as "Best Effort" (BE).

With the rapid growth of e-commerce, demand for various classes of Internet services is expected to grow and diversify. Customers with real time and business-critical data applications are searching for improved levels of service, or Quality of Service (QoS) connections to the WWW. Compared to BE, ISPs are now looking to premium QoS class connections for customers. These new classes could include, but are not limited to, a class to guarantee timely delivery, a class to ensure no traffic loss, a class for delivery confirmation, or any combination of all classes.

These developments are harbingers for the future. First, there will be at least two classes of service in the Internet market: BE and QoS. Since QoS includes BE service as its lowest class of service, the superiority of QoS to BE results in vertical product differentiation. Second, it is reasonable to expect a change in the pricing paradigm from non-metered, flat rate, unlimited access user pricing to usage-sensitive metered pricing for higher valued QoS.

ISPs are simultaneously competitors and cooperators. On one hand, they are competitors for market share. On the other hand, they are cooperators that provide universal, global connectivity. Thus, one ISP's actions influence another's actions. Furthermore, end-to-end QoS could not be established without strong cooperation among ISPs. To behave accordingly suggests game theoretic modeling, i.e., each player in the game is a competitor, and their interactions provide motivation for strategic decisions. Under traditional Internet pricing with unlimited access and flat rate monthly payments, users want to take as much bandwidth as they can within their access capacity. This leads to a "tragedy of the commons" phenomenon which can be overcome through usage-sensitive pricing. Therefore, we propose a simulation methodology to explain an ISPs' equilibrium behavior in a futuristic QoS market. To implement our approach, we employ a Bertrand price competition model to determine a customer's willingness-to-pay and Internet usage patterns. The Bertrand model also includes two different pricing schemes: one for an ISP with traditional unlimited access, flat rate pricing for BE, and another for an ISP with a two-part tariff consisting of a fixed rate for BE, plus a usage-sensitive pricing strategy for QoS.

The research is structured as follows. First, we present the differentiated service initiative of AT&T and WorldCom. Then, we provide a literature review forming a foundation for the research extension, followed by economic assumptions for pricing, consumer demand and usage for the industry. Next, cost, revenue and profit functions for the methodology are discussed. Finally, the proposed simulation and optimization approach, as well as the behaviors of ISPs at an equilibrium point are presented. To conclude, preliminary results and issues for future research are offered.

SUPPORTING INDUSTRY ACTION

Recently, large service providers like AT&T and WorldCom announced in the summer of 2001 that they would start providing Internet customers with "Class of Service" (CoS) connections using Multi-Protocol Label Switching (MPLS) and Differentiated Services. These CoS-based services consisted of four priority level classes: Platinum, Gold, Silver, and Bronze. Customers requiring applications for voice and video would probably choose the Platinum, or highest priority level of service, while other customers requiring only applications for HTTP and e-mail traffic might choose the Bronze, or lowest priority level of service (i.e. BE). Customers with business critical data applications might choose an intermediate priority level, namely Gold or Silver. However, these announced service levels were limited to connections completely contained in the carriers' own networks.

Also in 2001, the Florida Multimedia Internet Exchange (FMIX), managed by Bell South, announced a plan to be the first Network Access Point to use an MPLS interconnection among different providers. To do this, FMIX faced many new challenges with QoS interconnections, such as pricing, class matching between providers, as well as managing the disclosure of network information for end-to-end quality guarantees.

SUPPORTING LITERATURE

Our work builds upon recent research by Stahl, Whinston and Zhang as well as Gupta, Linden, Stahl and Whinston, where a simulation-based approach to a duopoly ISP environment was studied. Both studied the affects of BE, flat rate versus usage-based pricing. Stahl et. al (1998) found that when a company like AOL imposes a fee to maximize profits, dynamic usage pricing increased profits five times, while network-wide social benefits increased seven times. Gupta et. at. (2001) also showed that usage-based pricing enhanced system-wide benefits overall. We extend these research efforts to incorporate pricing guidelines for premium QoS choices for ISPs which could easily become a de-facto environment for the Internet in the near future.

In addition, we utilize Bertrand's competitive model to analyze what is considered to be an industry of narrow competition (Bertrand, 1883). Other researchers make use of a Cournot model to analyze the Internet industry, which is a reasonable assumption if a homogeneous service with limited capacity exists (Baake and Wichmann, 1988; Shin et. al., 2002a; Shin et. at., 2002b). Like Stahl et. al. and Gupta et. al., our model also assumes a duopoly in the Internet Access market. However, in our analyses, the service is not homogeneous and capacity is only briefly limited by its chosen queuing medium. We therefore believe that a Bertrand model better suits our approach.

Finally, it is generally known that when one firm's market penetration reaches approximately 60%, the market usually experiences price competition. This has been shown to exist in the radio, television and video cassette recorder markets. With this in mind, recent research indicates that U.S. online households were expected to surpass the 60% mark by the end of 2002 (Vanston, 2002). Hence, there is a strong probability that ISPs will wage a price war in the near future.

ECONOMIC CONDITIONS

In the following three subsections, information pertaining to the development of price, demand and usage functions is presented for model formulation. All are considered parameter inputs for simulation.

Pricing Functions

One important characteristic of the industry is that ISPs are competing with each other for market share. Thus, they are trying to maximize their own profits based on the belief that the other ISP's price is fixed. To model this behavior, we assume that there are only two service classes, a BE class and a premium QoS class. In addition, there is no quality difference among each ISP's QoS class, thus customers are indifferent as to whether they will consume QoS from ISP1 or ISP2. It is also assumed that there are two prices to enter the Internet access market:

(1) Access Price (F) for a right to connect to an ISP's network (a fixed price), and

(2) Usage Price (r) for the volume of Internet usage per hour (a variable price).

The price structure for each ISP can be expressed as:

(1) P1(F1, r1) for ISP1, and

(2) P2(F2, r2) for ISP2.

We further assume that ISP1 uses a flat rate pricing scheme, which can be reduced to P1(F1, 0), where the customers of ISP1 pay only $F1/month regardless of traffic type and volume of their connection hours, while the premium QoS of ISP2 incurs both access and usage price components.

Some have called this type of pricing inefficient because the added fixed charges may deter some users who, at marginal cost prices, would be willing to join the network and consume (Cawley, 1997). Cable television pricing refutes this charge. In that industry there is a fixed price to watch "basic" programming and an additional usage charge for high-valued programs that are handled on a "pay-per-view" basis.

The two-part tariff in our methodology has a similar form to the pricing scheme used by the Cable Television industry. The fixed part lump-sum fee is the right to use the lowest class of service, and the variable part is for the consumption of the premium class of service. Someone who only wants to use the "basic" service pays only the fixed part lump-sum fee. To obtain high-valued programming, the user must first purchase "basic" (BE) service to incur the premium "pay-per-view" (QoS) service.

Demand as a Function of a Customer's Willingness-to-Pay

To capture consumer demand, a recent United States General Accounting Office report was used. Its purpose was to study Internet usage. One of the questions asked was: "About how much do you pay per month to access the Internet from your home?" Although this question does not provide a customer's willingness-to-pay for Internet access, we can use the data as a proxy for consumer demand. Table 1 presents the distribution of respondents.

In addition, demand must be differentiated between BE service and premium QoS class. According to Gal-Or (1983; 1985), when a product is differentiated on the basis of quality, and each consumer is assumed to purchase only one unit of the product, the consumers' willingness-to-pay (W) is assumed to be dependent upon their taste factor (X) and a quality level (M) for the product. As a function, a consumer's willingness-to-pay takes the form:

W(X,M) = f(X)*M, where

[W.sub.X] > 0, [W.sub.M] > 0, [W.sub.MX] > 0, [W.sub.MM] [less than or equal to] 0, and [W.sub.XX] [less than or equal to] 0.

We further assume that the bandwidth needed for a specific class determines the quality level of that service, and we assume the bandwidth of the high quality class is at least twice as much as that of low class service, i.e, [M.sub.QoS]/[M.sub.BE] = 2. Therefore, in our model the willingness-to-pay for QoS ([W.sub.QoS]) is twice as much as that of BE service ([W.sub.BE]).

Customer Usage

The methodology proposed by this research also heavily depends on Internet usage patterns. To capture consumer usage, the same United States General Accounting Office report was consulted. One of the other survey questions asked: "On average, how many hours per week do you and all your members of your household spend on the Internet from your home?" Table 2 presents the distribution of respondents, which directly reflects usage data.

COST, REVENUE AND PROFIT FUNCTIONS

First, we know that cost is highly dependent upon the number of usage hours in each of the two classes. Previous research suggests that in the absence of price-based differentiation, users will choose the highest quality level regardless of traffic type (MacKie-Mason and Varian, 1995). Under this scenario all of the customers of ISP1 should choose the premium QoS class. However, according to parameter r2 of our model, customers of ISP2 choose a% for premium QoS class service and (1-a)% for BE class service.

Also, the cost structure of the Internet industry is characterized by a large, up-front sunk cost and near zero short run marginal cost. It is well known that, with a congestion-free network, the cost to carry or process an additional minute of Internet traffic approaches zero, because the incremental cost is near zero (Frieden, 1998). When an ISP provides for QoS traffic, he needs additional equipment, higher skilled labor, and must plan for a significant increase in operating cost (mainly for monitoring, billing and collection). Hence, we assume that each ISP incurs the same amount of equipment and human cost, so we do not include these two cost factors in our model. However, we have already assumed that the bandwidth requirement of QoS is twice that of BE service, but the cost difference between the two is far more than double. Considering a scaling effect, we propose $0.01 per hour as the cost of BE service, and $0.10 per hour as the cost for premium QoS. Formulations for the two ISP cost functions, ISP1 Cost and ISP2 Cost, are developed below.

(1) ISP1 Cost=$0.1/hr*4.3 wks/mon*S(ISP1's hrs/wk usage)

(2) ISP2 Cost=[($0.01/hr*(1-a)%)+($0.1/hr*a%)]*(4.3wks/mon*S(ISP2's hrs/wk usage)

Next, revenue functions for the two ISPs are developed. Each revenue function simply price multiplied by quantity.

(1) ISP1 Revenue = F1 * q1, where F1 is the unlimited QoS connection flat rate, and q1 is the number of ISP1's subscribers.

(2) ISP2 Revenue = (F2 * q2) + S([h.sub.QoS] * r2), where F2 is the fixed rate for the unlimited BE connection, q2 is the number of ISP2's subscribers, [h.sub.QoS] is the S(total connection hours of ISP2's subscribers)* a%, and r2 is the QoS connection rate per hour.

Finally, profit functions are easily developed as revenue minus cost. In our price model, we assume F1 to be higher than F2 since F1 covers both the fixed and variable pricing components. Each ISP's profit function is shown below.

(1) ISP1 Profit = ISP1 Revenue--ISP1 Cost, and

(2) ISP2 Profit = ISP2 Revenue--ISP2 Cost.

PROPOSED SIMULATION AND OPTIMIZATION

Simulation of the methodology begins by determining consumer demand. To do this, we employ a Random Number Generator (RNG) with the empirical distribution of Table 1 to obtain specific willingness-to-pay values. To determine consumer usage, we employ a two-stage RNG method based on the empirical distribution from Table 1 and the piecewise uniform distribution from Table 2.

Table 3 below summarizes the parameters and suggested ranges for the proposed simulation model.

Next, customers are assigned willingness-to-pay values ([W.sub.BE] and [W.sub.QoS]) along with a value for their Internet usage hours (h), which are generated by the RNGs described above. We assume that customers are aware of each ISP's pricing strategy, and that they choose their premium QoS provider to optimize their benefit. At the same time, each ISP also knows its competitor's pricing strategy and can construct the best choice among all known combinations of pricing strategies of ISP1 and ISP2.

With assigned values for [W.sub.BE], [W.sub.QoS] and h, each consumer is able to calculate his net benefits from the consumption of premium QoS, i.e., the difference between the willingness-to-pay for QoS and price of each ISP. For example:

(1) Net1 = [W.sub.QoS]--F1 by consuming QoS from ISP1, and

(2) Net2 = [W.sub.QoS]--(F2 + ([h.sub.QoS] * r2)) by consuming QoS from ISP2.

If both Net1 and Net2 are below zero, a customer will not buy from either QoS provider. If Net1 or Net2 is greater than zero, the customer will choose the ISP that will give him a higher net benefit value. Thus, if Net1 > Net2, the customer will choose ISP1, otherwise he will choose ISP2.

According to Bertrand's model, ISP1 will choose a price, F1, for its optimal profit assuming ISP2's price is fixed. ISP2 will also choose a price, F2 and r2, for its optimal profit under the assumption that ISP1's price is fixed. Output from the simulation calculates all possible profits of ISP1 and ISP2. The best response profit for each ISP is then chosen (a Nash equilibrium by definition).

In our methodology, we strive to find an optimal pricing strategy for total profit through access price competition (F1 and F2) holding r2 constant. We then iterate the process, each time using an increasing value for r2 until equilibrium occurs. Lastly, the final equilibrium point occurs at the intersection point of each ISP's best response function. Thus, by comparing ISP1 Profit [F1*, (F2, r2)] and ISP2 Profit [F1, (F2*, r2)], we find the simulation methodology's equilibrium at (F1*, F2*).

Trials of the simulation methodology have been easily conducted using the CSIM Simulation Package on the platform of Visual C++ 6.0.

CONCLUSIONS

Our proposed methodology indicates that an ISP's optimal QoS pricing strategy can be determined by the price of BE service along with a customer's preference toward the premium QoS option. Initial data indicates F1 = $30, and F2 = $10 with r2 = $1.10. Generally speaking, customers with small amounts of premium QoS usage prefer the two-part tariff. Conversely, customers with larger amounts of QoS usage prefer a flat rate pricing scheme. At what point and how we separate the two is what matters.

Many scholars and industry experts indicate that over-provisioning and traffic engineering cannot successfully provide premium QoS without an appropriate pricing scheme. Therefore, the introduction of usage-sensitive pricing into the Internet industry is probably inevitable. Unfortunately, our simulation methodology cannot predict, unconditionally, which pricing strategy would provide the best market position for an ISP in the future. The problem is too dynamic. As prices change and new competitors enter or leave the market, our methodology can be implemented to determine an optimal, short-term pricing strategy. As conditions continue to change, the process can be iterated with newer data to provide a more current strategy.

To conclude, this research provides a foundation for simulating pricing strategies in the ISP market. In future empirical research, we plan to incorporate more in-depth factors such as time-sensitive data, more choices for usage prices, and consumer taste and quality for ISP choice. Consumer taste and quality variables would provide us with other elastic willingness-to-pay factors. We realize that our current assumption of "2" for the willingness-to-pay factor is somewhat rigid. This, along with several of the factors listed above will certainly provide greater sensitivity analyses, and possibly lead to an improved methodology.

REFERENCES

Baake, P. & T. Wichmann (1999). On the economics of Internet peering. Netnomics, 1 (1), 89-105.

Bertrand, J.L.F. (1883). Theorie Mathematique de la Richesse Sociale. Journal des Savants, 67, 499-508.

Cawley, R.A. (1997). Interconnection, pricing, and settlements: Some healthy jostling in the growth of the Internet. In Kahin, B. & J. Keller (1997). Coordinating the Internet (pp. 346-376). Boston, Massachusetts: MIT Press.

Frieden, R. (1998). Without public peer: The potential regulatory and universal service consequences of Internet Balkanization. 3 Virginia Journal of Law & Technology 8, available at http: //vjolt.student.virginia.edu/.

Gal-Or, E. (1983). Quality and quantity competition. The Bell Journal of Economics, 14(2), 590-600.

Gal-Or, E. (1985). Differentiated industries without entry barriers. Journal of Economic Theory, 37(2), 310-339.

Gupta, A., L. Linden, D. Stahl & A. Whinston (2001). Benefits and costs of adopting usage-based pricing in a subnetwork. Information Technology and Management, 2(2), 175-191.

MacKie-Mason, J.K. & H.R. Varian (1995). Pricing congestible network resources. IEEE Journal of Selected Areas in Communications, 13(7), 1141-1149.

Shin, S., H. Correa & M.B.H. Weiss (2002a). A game theoretic modeling and analysis for Internet access market. International Telecom Society 2002 Conference, Seoul, Korea.

Shin, S., M.B.H. Weiss & H. Correa (2002b). Internet QoS market analysis with product differentiation theory. International Telecom Society 2002 Conference, Seoul, Korea.

Stahl, D. A. Whinston & K. Zhang (1998). A simulation study of competitive Internet pricing: AOL flat rates versus GSW usage prices. Proceedings of the first International Conference on Information and Computation Economies, 68-76.

USGAO (2001). Characteristics and Choice of Internet Users. United States General Accounting Office publication GAO-01-345.

Vanston, L.K. (2002). Residential Broadband Forecasts. A white paper by Technology Futures, Inc., available at www.tfi.com.

Sengjae Shin, Mississippi State University--Meridian

Robert F. Cope III, Southeastern Louisiana University

Rachelle F. Cope, Southeastern Louisiana University

Jack E. Tucci, Mississippi State University--Meridian

Table 1: Household Expenditure for Internet Access
(USGAO, 2001)

 $0 ~$5 ~$10 ~$15 ~$20
 8.9% 1.4% 3.8% 8.3% 21.0%

 ~$30 ~$40 ~$50 $50 an up
31.7% 11.1% 8.7% 5.1%

Table 2: Internet Usage Distribution (hrs/wk)
(USGAO, 2001)

 0-4 hrs 4-10 hrs 10-15 hrs 15-25 hrs
 6.3% 12.1% 19.4% 29.3%

25-40 hrs 40-60 hrs 60-90 hrs
 19.8% 6.3% 6.9%

Table 3: Suggested Simulation Input Parameters

 Parameter Suggested Range

[W.sub.BE] $0 to $50
[W.sub.QoS] $0 to $100
 F1, F2 $10, $15, $20, ..., $100
 a 20%, 30%, 40%, 50%, 60%, 70%
 r2 $0.3, $0.5, $0.7, $0.9, $1.1, $1.3
 h 0 to 387 hrs/mon (4.3*90)