5.6. Chronic Cyclicality in Employment and Production and How to Cure It
5.6.1. The Curious Effect of Random Variations in Demand
Return to the model called 'Production and Workforce'. Open the icon for the retail order rate at the top of the production control sector. You will see the following equation: Retail Order Rate = 1 000 * NORMAL(1,0) + STEP(100,10) + RAMP(0,10) {refrigerators/week}. This is a versatile formulation for the exogenous retail order rate that allows a variety of different demand assumptions to be tested. We have already seen in Figures 5.19 and 5.20 the simulated effects of a one-time unexpected increase in demand from 1000 to 1100 refrigerators per month. This step change led to fluctuations in production and employment reminiscent of cyclicality experienced in the real factory, except that the fluctuations gradually subsided. To inject more realism, we repeat the simulation with the addition of randomness in the retail order rate. In the equation for the retail order rate, rewrite NORMAL(1,0) as NORMAL(1,0.05), which means that demand is normally distributed around 1 000 with a standard deviation of 0.05 or 5 per cent. Meanwhile, the original step increase in demand still takes place in week 10.
Press the 'Run' button to obtain the time charts shown in Figure 5.24. The retail order rate (line 1, top chart) is highly variable from week to week, moving in a range between 1 000 and 1 200 refrigerators per week. The one-time increase in demand is masked by randomness. Nevertheless, and this is the interesting point, both the desired production rate (line 3) and the production rate (line 4) exhibit clear cyclicality, very similar to the deterministic pattern seen in Figure 5.19. Note, however, that the fluctuation persists throughout the entire 160-week simulation rather than gradually subsiding. Meanwhile, the average ship rate (line 2, top chart) clearly separates the uplift in demand from the random variations, just as one would expect from a process of information smoothing. Refrigerator inventory (line 1, bottom chart) also smoothes out randomness, so the remaining variation is cyclical. What we are seeing here is a vivid illustration of the principle that structure gives rise to dynamic behaviour. Even though retail orders vary randomly, the factory nevertheless finds itself locked into cyclical production and employment. Moreover, this instability will persist indefinitely if there is randomness to provoke it (for a more technical explanation of randomness and cyclicality in information feedback systems see Forrester 1961, appendix F on noise). The structure of asset stocks, flows, operating policies and information means that the factory is prone to chronic cyclicality – entirely consistent with the problem observed in practice. You could say the factory's mystery is solved.
5.6.2. Industry Cyclicality and Business Cycles
The cyclical dynamics of the factory model are not confined to individual firms. There is evidence that similar processes of inventory control and workforce management are responsible for unwanted cyclicality in entire industries and in the economy as a whole (Mass, 1975). Figure 5.25 shows inventory coverage in US manufacturing industry over a period of 50 years from 1950 to 2000. (The data does not include finished inventories held by retailers and other distributors outside the manufacturing sector.) Coverage is on a scale from 1.3 to 2 months. The trajectory is strongly cyclical with a period of 5–8 years. This erratic pattern is typical of cyclicality in the real world where randomness interacts with feedback structure to cause fluctuations of varying periodicity. It is interesting to note the behaviour between 1990 and 2000 when inventory coverage falls far below its long-term average. Various explanations of this phenomenon are possible. It could be that widespread use of information technology and just-in-time systems enabled firms to operate with less inventory. Another explanation is that 1990–2000 was a period of sustained economic growth, accompanied by chronic inventory shortage.
Figure 5.24. Simulation of a 10 per cent increase in demand and five per cent random variation. The view in production control
Cyclicality also occurs in financial services. Figure 5.26 shows the insurance underwriting cycle between 1910 and 2000. Despite profound changes over these 90 years the cycle has been remarkably consistent. Underwriting profits, expressed as a percentage of earned premiums, fluctuate sharply although there are no physical inventories and materials in the supply chain. The detail business processes and procedures that create this cycle obviously differ from those in manufacturing, but somewhere among them are analogies to forecasting, inventory control and workforce planning that create a balancing loop with delay.
Figure 5.25. Cyclicality in US Manufacturing Industry
Source: Sterman, J.D., Business Dynamics: Systems Thinking and Modeling for a Complex World, © 2000, Irwin McGraw-Hill, Boston, MA. Reproduced with permission of the McGraw-Hill Companies.
Figure 5.26. Cycles in Service Industries
Source: Sterman, J.D., Business Dynamics: Systems Thinking and Modeling for a Complex World, © 2000, Irwin McGraw-Hill, Boston, MA. Reproduced with permission of the McGraw-Hill Companies.
5.6.3. Policy Formulation and What-ifs to Improve Factory Performance
How might the factory's performance be improved? Such a question is natural at this stage of modelling, and brings us to the topic of policy formulation and what-ifs. What changes to operating policies might alleviate cyclicality? Various ideas come to mind and these ideas can all be tested with simulation. We know that cyclicality arises in the balancing loop that links inventory control, workforce management and production. What if we changed the policy for inventory control, say by halving the normal inventory coverage or doubling the time to correct inventory? What if we made the factory more responsive by slashing the workforce planning delay and trimming the time to adjust the workforce? All these are testable changes within the framework of the model. To illustrate, try doubling the time to correct inventory from eight to 16 weeks. Why this particular change? We know from earlier simulations that the process of rebuilding inventory tends to amplify demand changes, creating a misleading and exaggerated impression of desired production and desired workforce. Hence, if rebuilding were spread over a longer period of time, it may be easier for the factory to plan its workforce.
Figure 5.27 is a simulation of the new, more relaxed, inventory control policy in response to a 10 per cent increase in demand, without randomness. There is a noticeable improvement in the stability of production and inventory by comparison with Figure 5.19. Desired production (line 3, top chart) peaks at 1 187 refrigerators per week in week 30 compared with a peak of 1250 in Figure 5.19. Refrigerator inventory (line 1, bottom chart) falls to a minimum of 2 838 in week 30, slightly lower than the minimum of 3 000 in Figure 5.19. The factory is more willing to tolerate an inventory imbalance and because of this new attitude there is less short-term pressure to rebuild inventory. The knock-on consequence is a more gradual rise in desired production that translates, through the workforce, into more stable production. The production rate (line 4) rises to a peak of 1 164 in week 45. There is much less amplification of production relative to retail orders and so the subsequent overbuilding of inventory in the period between weeks 60 and 75 is reduced, enabling the factory to achieve more easily a proper balance of supply, demand, inventory and workforce.
Figure 5.27. Simulation of a 10 per cent increase in demand when the time to correct inventory is doubled from 8 to 16 weeks. The view in production control
The essence of the policy change is for the factory not to overreact when inventory is too high or too low. Strong inventory control sounds good in principle but leads to unnecessary changes in the workforce that eventually feed back to destabilise production and inventory. Patience is a virtue in situations where corrective action depends on others. The same was true of the imaginary hotel showers in Chapter 4 where two shower-takers try to stay warm while unknowingly sharing hot water. It is better to be patient when there is interdependence. However, if local corrective action is genuinely possible, without delay, then it can pay to act swiftly, just as intuition would suggest. This phenomenon can be seen in the factory model by halving both the workforce planning delay and the time to adjust workforce. For easy comparison, the time to correct inventory is restored to eight weeks. So this is a factory with strong inventory control and a flexible workforce. The simulation is shown in Figure 5.28. The flexible factory avoids unnecessary fluctuations but is also faster to respond than the patient factory. By comparison with Figure 5.27, production (line 4, top chart) builds more quickly and inventory (line 1, bottom chart) falls less in the aftermath of the same uplift in demand. Moreover, the factory achieves equilibrium sooner. One conclusion is that a flexible factory can outperform another factory that is slow to adjust workforce (or capacity more generally). Perhaps even more important, however, is the observation that a patient factory can perform very well, much better than the base case, without the need for workforce flexibility. Success is achieved by ensuring that decisions in production control take into account the inevitable delays in planning and adjusting the workforce.
Figure 5.28. Simulation of a 10 per cent increase in demand when the workforce planning delay and time to correct workforce are both halved. The view in production control
