Green Computing as a CSR Real Option
Online Appendices
Appendix 1: Modeling Profit-Driven CSR Investment on Virtualization as a Bayesian Real Option
Energy Usage Profile
An energy usage profile, which measures ongoing energy consumption cost and environmental impact, is fundamental to both the design of application architecture and the evaluation of green computing alternatives. In the evaluation of green computing alternatives, an annual energy usage profile should be developed over the entire economic life of the project, not just over the initial year. When measuring the effectiveness of application architecture, both energy consumption and environmental impact should be considered (Chheda et al. 2008). For any year , the energy usage profile () can be expressed as the following relationship, with () as the energy consumption cost and () as the energy impact:
(1)
For period, suppose is the energy consumed by computing equipment in watts; is the electricity charge in dollars per kilowatt-hour (kWh); is the number of hours the equipment is used per day; is the number of working days in a year; is the carbon tax in dollars per short ton of CO2 emissions; and is the quantity of CO2 emitted to the atmosphere in lbs. per kWh. Based on these variables, the annual energy consumption cost can be expressed as, with total energy consumed in kWh as and the annual cost of CO2 emissions as , since there are 2,000 lbs. of emissions per short ton.
Real Option Model on Profit-Driven CSR with Carbon Tax Uncertainty
Next, we formalize the CSR investment to implement a green computing alternative as a real option on carbon tax uncertainty. The green alternative is to replace an existing server system (System A) with a more energy-efficient server system (System B). We assume that there is uncertainty in the carbon tax, as is the case currently in Ontario, where details of the cost and timing of implementation are not yet known. Furthermore, there is uncertainty over the degree to which the carbon tax will be enforced and whether all businesses or only certain industry sectors will have to comply. Therefore, when undertaking CSR investments to “go green,” as in the above example, organizations would not immediately make the CSR investment unless there is an economic benefit. An organization may forgo making such an investment until more information is available. In the above setting, we can compare the CSR investment in green technology with an option to wait for more information (or the real option to delay the investment). Sanders et al. (2013) provide an example that is similar to the green computing setting. In their example, a company will delay an investment in CO2 capture and storage despite societal interest in preventing global warming.
If an organization can obtain additional information to resolve uncertainty surrounding a carbon tax by postponing the CSR investment, then it is important to consider the trade-off between the cost of obtaining the additional information and the economic upside value of waiting. The issue of obtaining more information and resolving uncertainty can be investigated by integrating Bayesian analysis into the CSR option of whether or not to convert from System A to System B.
Sources of Uncertainty
The main sources of uncertainty in the CSR options are uncertainty about the level of carbon tax, uncertainty in the degree of enforcement of the carbon tax, uncertainty in the hardware cost of the new server system (due to new technologies that will lower prices of IT equipment), uncertainty about electricity prices, and uncertainty in estimating the staff hours associated with the new server system. In order to simplify, we assume the last two sources of uncertainty to be nonexistent and electricity prices and staff hours to be constant throughout the life of the green computing investment project.[1] Uncertainty in the carbon tax arises because, at the time of undertaking the CSR investment, the firm is unable to precisely forecast the carbon tax that will be levied. More specifically, there is imperfect information regarding the amount of carbon tax. In the proposed model, uncertainty in carbon tax is modeled using states of nature (or a discrete probability model), as in a decision tree analysis framework. Asset price uncertainty (that is, uncertainty concerning the present value of the cash inflows) is a standard feature in most real options (Grenadier and Malenko 2010). Therefore, uncertainty pertaining to a carbon tax can be considered to be similar to asset price uncertainty because cash inflows are a function of the impact of the carbon tax. The uncertainty in carbon tax results in the intrinsic value of the CSR real option at time zero because the environmental impact (EI) increases the terminal value of the CSR real option at each state (the low carbon tax state and the high carbon tax state).
Another critical source of uncertainty is the degree of enforcement of the carbon tax. As Grenadier and Malenko (2010) discuss in their article on the Bayesian approach to real options, cash flow uncertainty pertaining to future shock (asset price uncertainty) may not be the only type of uncertainty in real option models. They argue that uncertainty related to whether future cash flows are permanent or temporary, which they label as past shocks, is also critical. Firms are able to resolve this uncertainty by Bayesian updating or by learning about the nature of past shocks over time by delaying an investment. Thus the evolving uncertainty is not constant in their continuous-time Bayesian model and is driven by Bayesian updating, resulting in two real options, a traditional “waiting” option and a “learning” option. The latter allows the manager to learn more about the nature of past shocks (Grenadier and Malenko 2010). In the CSR option in green computing, the firm cannot identify the exact nature of the degree of enforcement of the carbon tax. Therefore, in addition to the uncertainty in the carbon tax, we consider the uncertainty in the degree of enforcement that results from political and industry pressure.
In order to resolve the combined uncertainty, the firm can defer undertaking the CSR investment by a year and obtain additional information, which results in a CSR option to delay the green computing investment. Therefore, there are two value components (real options) in the proposed CSR options framework consisting of (1) the intrinsic value of immediately exercising the real option to exploit the upside potential of the incremental dollars saved by minimizing the environmental impact of using the new server system and (2) the real option to defer undertaking the CSR investment to obtain additional information, the “delay option.”
The third source of uncertainty is the hardware cost of a new server system that results from postponing the CSR investment by a year. Typically, technology innovations lower the cost of product components and thus the price of servers. For example, Koomey et al. (2009) point out that the assumption of cost reduction and growth in computing speed related to Moore’s law would be true at the level of individual server systems, while energy costs tend to increase. Therefore, modeling hardware cost uncertainty would involve treating the exercise price of the CSR real option as reducing over time. The exercise price due to uncertainty in the server costs can be modeled in two ways. We discussed this modification in depth in the next subsection.
Definition of Model Parameters
In order to formulate the CSR investment as a real option to postpone investing in enterprise architecture until more information is obtained, we define variables pertaining to the green computing alternatives as follows.
Exercise Price of the Real Option
In the context of a real option, the CSR investment cost is the exercise price of the option. There are several elements which need to be considered. They include hardware costs of the new server system (), onetime initial software license savings (transfer of existing licenses) due to new servers using fewer software licenses (), and migration costs of new servers consisting of staff costs for planning and consolidation (). Another element to consider is the net cost or profit from the disposal of the old server system. More specifically, when a new server system (System B) is considered at time zero, either the old server system (System A) may be sold for salvage value or the organization may incur an environmental cleanup cost for waste disposal. These salvage values are treated as onetime cash inflows at the time of disposal, and they are subject to tax considerations.[2] The environmental cleanup cost for waste disposal is treated as a onetime cash outflow in the analysis. Suppose is the salvage value of the old server system and is the respective environmental cleanup cost for waste disposal. Then in modeling the net cash flows on disposal of old servers, one of the following two mutually exclusive alternatives arises (sell System A or dispose of System A as waste). We illustrate the alternative to sell System A by considering as a onetime cash inflow to be included in the exercise price.
The opportunity cost of space savings from virtualization can be modeled as follows. Let be the number of old type A servers, and let be the number of new type B servers that will replace A. Since, the reduction in the number of servers is . Suppose each server rack can hold number of either type of servers. Then the number of server racks reduced by replacing System A with System B is . Let and be the length and width of the base of a server rack, and let be the walkway allowance for the front and back of a server rack (all measured in feet). Then the floor area per server rack can be estimated as . The total floor area saved as a result of the new server system can be estimated as . The freed-up floor area due to virtualization can be alternatively converted, for example, into office space. If is the office rent charge per square foot, then the opportunity cost of the freed-up space due to virtualization if used as office space is . This opportunity cost will reduce the investment cost and thus should be included as part of the exercise price.
Finally, the data storage cost can be modeled as follows. Let be the storage capacity of a disk array in terabytes, be the operating system disk storage required by each virtual machine in gigabytes, and and be the number of virtual machines that a type A server and a type B server can support (i.e., ). Then the number of storage disk arrays required to support type A servers and type B servers can be estimated as and . Note that each storage disk array can be connected to servers of type A or servers of type B using two network switches (two switches are used, one for backup in case of a failure). Let be the cost of a network switch and be the cost of a storage disk array. If the old storage disk arrays have to be entirely replaced, then there will be an additional data storage cost pertaining to the new server System B equal to . The data storage cost has to be added to the exercise price. Alternatively, if existing storage disk arrays of the old server System A can be reused, then there will be a reduction in the number of data storage disk arrays required for the new System B. This reduction in the number of storage disk arrays is estimated as and there will also be network switches that are not required. These redundant storage data arrays and the network switches can be kept for future use or may be sold for salvage value. If they are sold, then the cash inflow from the sale has to be considered as part of the exercise price. Thus the exercise price of the CSR real option or the capital expenditure at time zero is.
According Moore’s law, new technology can lower the price of servers during the time period the CSR investment is delayed. Consequently, the exercise price of the CSR real option can decrease over time, which is usually the norm in IT capital investments (Koomey et al. 2009). Suppose the exercise price (investment cost) due to the reduced cost of the new server system decreases in a nonrandom way. In the decision tree approach, this problem can be easily incorporated by changing the exercise price in each state to the new exercise price (Copeland and Antikarov 2001). More specifically, if the cost of a new server a year from now is dollars and is lower than what it was at time zero (), then the new exercise price in each state at time is .
If, however, the reduction in new server cost is assumed to be random and to change in a stochastic manner, then the uncertainty in the exercise price can be modeled by considering the joint uncertainties of the new server cost and carbon tax. In order to model the joint probability of both uncertainties, it is assumed that the two sources of uncertainty (server cost and carbon tax) are independent. More specifically, the sources of uncertainties arrive sequentially during the delay period. Uncertainty in carbon tax occurs in the first six months, and the uncertainty in new server cost occurs during the next six months. For the uncertainty pertaining to the carbon tax, there will be the two states of nature: a high carbon tax with probability and a low carbon tax with probability. Suppose uncertainty regarding new server cost is also modeled using two additional states of nature “new server cost-1” with probability and “new server cost-2” with probability. From a discrete modeling point of view, the sequential arrival of two types of uncertainties will result in four states of nature at the end of year one: (1) high carbon tax, new server cost-1; (2) high carbon tax, new server cost-2; (3) low carbon tax, new server cost-1; and (4) low carbon tax, new server cost-2, with joint probabilities (,,, ).[3]