Table of Contents
This research is funded by an Australian Research Council large Discovery Project grant in 2006-2008. The research started with investigations into the entry by new firms into the Flat Panel Display industry, and the discovery that all successful entries were clustered in downturns; not a single firm successfully entered the industry during an upturn. This research thus established a strong link between strategizing by firms and the cyclical industrial dynamics of the industry.
Business Cycles and Industry Cycles
Research into business cycles has established a large body of literature. Apart from early efforts that sought to attribute business cycles to exogenous factors such as periodicity in sun plots (W. S. Jevons, 1835- 1882), or the planet Venus (H. L. Moore, 1869- 1958), the most influential work in the area is perhaps Joseph A. Schumpeter’s 1939 masterpiece, Business Cycles: A theoretical, historical and Statistical Analysis of the Capitalist Process. This work elaborates on his early work expounded in The Theory of Economic Development (1912).
Four major business cycles have been intensively discussed by Schumpeter and others: the Kitchin cycle (3-5 years); Juglar cycle (7-11 years); Kuznets Cycle (15-25 years) and Kondratieff cycle (45-60 years). There is by now an extensive literature on these macro-business cycles.
Our concern however is with industry-level cycles. These are meso-level phenomena, between the micro-level of individual firm behavior and the macro-level where business cycles can be identified and measured. The business cycles at the macroeconomic level are in fact a reflection and aggregation of microeconomic structure and behavior. But there are at least two features that distinguish between the ‘business cycle’ and the ‘industry cycle’.. First, multiple time series are examined for identifying business cycles, including real gross domestic product, unemployment rate, personal income, index of industrial production; and so on. Some of the indicators may be much less relevant to individual industries than others. Second, highly aggregated data are employed. Heterogeneity across industries is therefore ignored and missed.
Cyclicality of an industry can not only create remarkable opportunities for new entrants, as revealed by the entry behavior of firms in Flat Panel Display industry; but also can have substantial implications for other strategic decisions by firms, e.g. firms' capital / R&D expenditure decision. This theme can be illustrated by the semiconductor industry and specifically by the behaviour of Intel during the recent downturn in the semiconductor industry cycle.
The downturn of the global semiconductor industry cycle in 2000-2002 was perhaps the worst in the history of the industry. When prices were falling, production was falling, and revenues were falling, not to mention profits falling, this was the time that Intel chose to ramp up investment. Intel invested in new fabrication facilities that would be 250% more productive than those of their competitors as the industry swung into its expected upturn. But Intel was not just investing in new fabs that could produce chips on 12-inch wafers at 90 nm line widths. Intel was diversifying into new fields, finding new applications for new chips. On top of its success with WiFi chips, which give consumer products and communication devices like cellphones inter-operability over a short span (up to 200 feet), Intel in 2002 introduced new WiMax chips, which extend the range of radiofrequency devices to 30 miles – enough to cover a small city.
How do we characterize Intel’s strategizing during this downturn of 2000-02? Intel was not an oligopolist capturing rents from its rare resources (as the resource-based view may argue) or from its privileged market position (as Porter’s competitive forces view may suggest). As Business Week described it, Intel was engaging in massive counter-cyclical investment during the downturn, using the period when its competitors were scaling back R&D, capital expenditure and production as a time to build new product lines in new digital businesses as well as new fabrication facilities, to be ready when the next upturn comes. Its strategy is essentially time-sensitive, and this could thus be best viewed through the lens of the industry cycles.
Cyclical industrial dynamics have profound implications for the way that firms strategize around the industry, whether they be incumbents or contemplating entry. Their significance for strategy is illustrated by the Intel case of counter-cyclical investment during the 2001-2 downturn, but many other cases could be investigated, including the behaviour of Samsung in the memory chips (DRAMs) sector of the semiconductor industry, and the strategic behaviour of firms in other industries such as Flat Panel Displays.
2. Cyclical Industrial Dynamics in a Single Industry
The first issue to be explored in this research project is the character of cyclical dynamics in a single industry. Five research questions are raised: (1) Can the turning points of the cycles be precisely determined? (2) How has the industry been driven by the aggregate economy? (3) Can we identify the relationship of the cycles across industries? (More specifically, do the cycles of a downstream industry ‘cause’ those of the upstream industry?) (4) Can we detect the underlying 'cyclical components' from one time series that exhibit cyclicality? (5) And finally, what are implications of industry cycles for firms’ strategic behaviors?
By taking the global semiconductor industry as a case-study, we have been able to utilize a number of advanced analytical tools, including the NBER Cycle Dating approach, the Vector Autoregression Model and the Fourier analysis, and apply these to the real industrial data.
Dating turns for business cycles has long been practiced. However, industry cycles have rarely been dated with relatively rigorous methods. In business cycle research, the numerous dating methods can be grouped according to whether the method is ‘parametric’ or ‘non-parametric’. A parametric method develops statistical models such as cosine and sine waves which fit data and select turning points with the estimated parameters. By contrast, a non-parametric method establishes a set of dating rules and identifies turns in a ‘model free environment’.
We apply the so-called the Bry & Boschan method which has long been used as a ‘conventional approach’ to cyclical analysis by NBER (the US National Bureau of Economic Research) to the semiconductor industrial data. The cycles resulting from the process are compared with those suggested by industry experts such as consultants’ reports like IC Insights. We establish that overall the two sets of 'cycles' are largely consistent with each other, suggesting the validity of the cycles identified.
Just as white light can be separated into spectral components with different wavelengths, we ask whether underlying cyclical components can be detected in an industrial time series. By using Fourier analysis, we identify three pronounced cyclical components in the data series, with wavelengths of 4 years, 2.29 years, and 1.03 years accordingly. The model built on the three cyclical components fits the data well.
The Fourier analysis applied to the semiconductor industry data suggests multiple sources of the cyclicality. For example, the one-year cycle seems to correspond to the seasonal cyclical factor and the four-year cycle to the macroeconomic business cycle. The length of the latter is particularly consistent with the Kitchin inventory cycle. If these hypotheses are true, then the outstanding cyclical component with a length of 2.3 years, as we suspect, may correspond to one or more industry-specific factors. This would be the semiconductor industry cycle proper.
Apparently the aggregate economy, or the general business cycle, effects an industry's performance. Besides that, we expect a number of industry-specific factors to make their presence felt. The impacts on cycles of a particular industry from upstream industries and downstream industries in a value chain are evident. By building a Vector Autoregression model, we established that the cycles of the global semiconductor industry and the global PC industry are mutually reinforced. In other words, there is a reciprocal or 'bi-directional' relationship along the evolution of the two important industries.
3. Interaction and Co-movement of Cyclical Industrial Dynamics in Multiple Industries
More information available soon.
4. Strategizing with Cyclical Industrial Dynamics
More information available soon.
5. Publications and Working Papers
Mathews, J. A. 2005. Strategy and the Crystal Cycle. California Management Review, 47(2): 6-32.
Tan, H. & Mathews, J. A. 2007. Semiconductor Industry Cycles. DRUID-DIME Academy Winter 2007 Conference. Aalborg, Denmark.
Professor John A. Mathews holds the Chair of Strategic Management at Macquarie Graduate School of Management, Sydney. He has been a member of the Faculty at Macquarie since June 1998, where he teaches courses in Strategic Management, Competition and Strategy in Asia-Pacific, and Global Strategic Management, in Sydney, Singapore and Hong Kong. He served as Director of Research at MGSM from 2000 to 2004 responsibility for doctoral students in the PhD and DBA programs.
If you would like to contact Professor John Mathews via e-mail, click here
Hao Tan works full-time with Professor John Mathews primarily on the research project of cyclical industrial dynamics. He is also a Doctor of Business Administration Candidate at MGSM under the supervision of. Professor Mathews. Prior to joining the School in 2003, he was a management consultant in China. Before that he worked as a corporate planner at Huawei Technologies, an emerging Chinese player in the global telecommunication equipment industry. Hao earned both his M.B.A and B.S. in China.
If you would like to contact Hao Tan via e-mail, click here
If you wish to comment on this article, please email either John A. Mathews or Hao Tan. Their email addresses are:
John A. Mathews: John.Mathews@mgsm.edu.au
Hao Tan: Hao.Tan@mgsm.edu.au