Efficiency Is Fragility

A twentieth-century trade statistic assumed a good was made in one country and sold in another. That assumption stopped being true somewhere around 1990, and the statistics have never fully caught up.

Containerisation dropped the cost of shipping a box across an ocean to nearly nothing. Cheap communication made it possible to coordinate production across time zones. Together they made it viable to slice manufacturing into stages and put each stage wherever it is cheapest: design here, chips there, screen somewhere else, assembly somewhere else again.

The consequence for the numbers is that a component crossing a border is counted at its full value every time it crosses. An assembled phone leaving China is recorded as a Chinese export at its whole value, although most of that value was added elsewhere and China's share is the assembly margin. This is why trade-in-value-added statistics had to be invented, and why bilateral deficit figures (much beloved of politicians) are close to meaningless.

A country can run a deficit with one partner and a surplus with another and be exactly balanced overall. The overall current account tells you something real about saving and investment. A bilateral balance, taken alone, mostly tells you where final assembly happens.

The second thing that changed is inventory, and here is where the interesting trade-off lives.

Inventory is capital sitting in a warehouse doing nothing. Just-in-time production, developed at Toyota, treats it as waste to be eliminated: components arrive as they are needed, the warehouse shrinks, and the capital is freed. It raises returns substantially and it is one of the great operational innovations of the century.

It also removes the shock absorber.

A supply chain holding six weeks of stock can absorb a six-week disruption invisibly. A chain holding none transmits the disruption immediately and completely to everything downstream. The efficiency and the fragility are not two properties in tension; they are the same property, observed under different conditions. Slack is what you pay for the ability to survive surprises, and an optimiser will eliminate it because the surprises have not happened yet.

The pandemic and the Ever Given's week in the Suez Canal both demonstrated this at scale, and neither was a failure of anybody's planning in the ordinary sense. They were the bill for a decision that had been paying out for thirty years.

There is also a structural amplifier worth knowing about, because it explains a great deal of industrial volatility.

The bullwhip effect: small fluctuations in consumer demand produce progressively larger swings in orders as you move up a supply chain. A retailer seeing sales rise 5% orders 10% more, to rebuild stock and cover the expected trend. The wholesaler, seeing orders up 10%, orders 20% from the manufacturer. The manufacturer orders 40% more components.

Each tier is behaving sensibly given the signal it receives, and the signal is orders rather than demand. This is why upstream industries (chemicals, components, capital equipment) are wildly more volatile than the consumer demand they ultimately serve, and why sharing actual point-of-sale data along a chain, rather than only orders, reduces the amplification substantially.

Which brings us to the chip shortage, the clearest recent illustration of everything above.

Early in the pandemic, car manufacturers forecast a collapse in demand and cancelled chip orders. Semiconductor fabs reallocated that capacity to consumer electronics, where demand was booming. Then car demand recovered faster than anyone expected, and the manufacturers tried to reorder.

Semiconductor lead times run to many months and fab capacity takes years and billions to build. There were no alternative suppliers, because advanced chip manufacturing is concentrated in a handful of facilities. Car production worldwide was halted by a component costing a few dollars.

The vulnerability was concentration in one specialised node, and the general lesson is that a deep chain is only as robust as its least substitutable link.

The policy response has been reshoring and stockpiling, and it is worth describing honestly. Redundant capacity, buffer stocks and domestic production cost money continuously, in exchange for protection against events that are rare and severe. That is insurance, and insurance can be an excellent purchase (for medical supplies, food, and arguably semiconductors) provided it is understood as a premium rather than sold as a free improvement.

It is also worth noting that domestic production concentrates risk geographically rather than removing it. A country whose entire supply comes from one domestic plant has swapped a foreign single point of failure for a local one, which is a different risk rather than a smaller one.