You might suspect that the results in my previous email look good only because the consumer demand is constant. Well, I had the same doubt, so I checked what the randomness of consumer demand does to my previous results.

Figure 6 shows the effect of sensitivity coefficient on the cost for constant consumer demand and random consumer demand. As you can see, they look similar except that the introduction of randomness smoothes out the curves a bit.

Figure 6:

The randomness of consumer consumption does increase the global minimum cost (See Figure 7). Naturally, we no longer observe the (almost) zero cost.

Figure 7

So, what's happening to the behaviors of the inventory levels when the cost is minimized?

Let's take a look:

Figure 8

As you can see, there is no "laminar flow" any longer. But I see that the curves are nicely synchronized - well, kind of.

OK, so far so good. The fluctuations seem to be nicely repressed. But this is not the end of the story - something happens after Week 500. See the next figure.

Figure 9

Around Week 700, something happens to this supply chain and the system becomes unstable out of the blue. (I checked the consumer demand around that time, but it "appears" normal.) In other words, this catastrophic behavior merely "emerged" (SFI's favorite!).

The good news is that all agents/nodes don't panic and keep the same rule--as a result, this unexpected & volatile fluctuation stops around Week 950, and things go smooth again as if nothing had happened (See Figure 10).

Figure 10