Jamal Munshi, Sonoma State Univesity, 1991
Empirical evidence of the impact of information systems on the business enterprise has eluded researchers and continues to occupy the highest priority in the MIS research community (Miller 1990). Controlled experiments of the effectiveness of certain Decision Support Systems in assisting in the subjects to solve a given business problem (such as that of King and Rodriguez 1978) have had some success. But what is gained in in experimental control is lost in the ability to generalize the findings to actual field situations. Besides, the type of business problem posed and the measures of success themselves may be viewed as arbitrary. On the other hand, ambitious cross sectional field studies that have sought a measurable economic impact of information systems have been largely inconclusive possibly due to the low statistical power in the face of overwhelming and uncontrollable intervening and concomitant variables.
The real problem with large conventional cross sectional studies is definitional. What, after all, is an "information system" and what kind of impact is it expected to have? A payroll system, an ATM system, a spreadsheet sales forecasting model, and a stock market trading system are all "information systems" with little more in common than digital logic and human users. It is such broad definitions that have given us the 'dependent variable problem'. If these are all information systems, what is the single appropriate response variable to be sampled as a measure of MIS impact that would apply to all of them?
The difficulty of the dependent variable problem has driven many researchers to study the human user instead of the system itself. The upshot of this approach has been a large number of studies on 'user information satisfaction'. However, these studies have been criticized (Melone 1990) for avoiding the real question and for ignoring well known canons of organizational behavior that do not support the assumed connection between user satisfaction and system effectiveness. In order to overcome what we feel have been the obstacles to the identification of objective measures of MIS effectiveness, we present what we consider to be a new and practical approach to measures of system impact and effect and demonstrate its application in an empirical setting.
The 'case study' method of MIS research espoused by Lucas (1974), Markus (1983), Allen (1988), and Benbasat (1987), is used in this study to overcome these problems in assessing the impact of information systems. First a theory of the impact of information systems is formulated. Then unambiguous and testable implications of the theory are deduced for a specific MIS implementation. Then if 'fortuitous' conditions allow, these implications are tested in a field setting using quasi-experimental methods (Cook and Campbell 1979).
The field setting enhances realism and generalizability of results when compared with controlled laboratory experiments. The targeting of specific systems overcomes the problems related to the identification of the appropriate response variable. This study has targeted exchange automation systems and uses tests on a specific implementation of electronic trading in the NYSE to test the implications of the theory that market automation changes the price discovery process and therefore the price behavior.
Extant economic theories of markets and market microstructure may be used to formulate a theory of of the price effects of exchange automation. Studies by financial economists such as Barnea (1974), Beja (1979), Amihud and Mendelson (1989) and others indicate that trading mechanisms affect price behavior especially with regard to short term components of price volatility. Further, theories of market automation proposed by Garbade and Silber (1979), Amihud and Mendelson (1990), Schwartz (1985), Miller (1989) and others all indicate that exchange automation changes the microstructure of the market. Based on these studies we deduce and test the theory that automation changes the component of price volatility that is generated by the trading mechanism.
To test this theory we study the implementation of the Electronic Display Book on the floor of the New York Stock Exchange and postulate that the excess intraday price volatility that may be ascribed to trading mechanism will change when the EDB is deployed. Fortuitously, the EDB was deployed over a three year period using over 100 implementation dates to convert about a thousand stocks to the new technology. This allows the use of 'event time' methodologies to cancel out concomitant historical effects from the quasi-experimental design. Further, economic effects are factored out of intraday changes by using a regression technique which leaves residuals, termed 'excess intraday volatility'. This quantity is ascribed to trading mechanism effects and is therefore postulated as the appropriate response variable for measuring the impact of the information system.
Using trading volume and market movement as control variables, it is found with 'intervention analysis' that at trading volumes higher than 100,000 shares per day and in cases where the daily price movement is at least three ticks, the excess intraday price volatility of stocks is lower after EDB implementation (see Figure 1 below). At lower trading volumes or when the price is not moving (by three ticks or more), no difference in excess volatility is detected.
Volatility is an important property of markets. In particular, lower volatility is a desirable property in market design. This is because in a mean-variance world of risk averse utility maximizers governed by the efficient market hypothesis (Markowitz 1959, Sharpe 1963, Fama 1970) investors will demand additional returns to bear the excess volatility generated by the market mechanism. As a result the overall investment into productive assets in the macro economy would be less than it would have been had the stock market been frictionless. However, the only real function of markets is to provide liquidity and liquidity is related to volatility. Market design that lowers volatility is desirable only to the extent that this does not adversely affect liquidity. A test on liquidity is therefore performed and it is found that there is no evidence that at least one measure of liquidity is lower after the deployment of the electronic limit order book. We may therefore conclude as follows: The implementation of the electronic trading system has had a measurable impact on price behavior. Volatility has been lowered without also decreasing liquidity. The 'quality' of the EDB market is therefore considered to be higher than that of the manual limit order book market.
For empirical research the problem of low statistical power may be addressed by tightening the functional definitions whose success is to be directly measured. For example, it may be difficult to measure the direct effect of a new accounts receivable system on the wealth of the shareholders but the measurement of the effect on mean collection time and percentage of the Receivables actually collected within a given time frame could be attempted. Such a measure can then be used to assess the ultimate effect on firm valuation using accounting methods.
Original file name: nyse
This file was converted with TextToHTML - (c) Logic n.v.