Total factor productivity is the economist’s name for “we cannot explain this part.” You measure labor, capital, and raw materials going in. You measure output coming out. The gap between what the inputs predicted and what actually happened is TFP. High TFP means something multiplied the inputs beyond their individual contributions. Low TFP means the inputs underdelivered. TFP is a synergy measure disguised as a technology measure.
Simple Picture
You give two chefs the same ingredients. One produces a decent meal. The other produces something extraordinary — the flavors combine in ways the ingredient list cannot explain. The difference is not “more ingredients” or “better ingredients.” It is the interaction between them. TFP is the economic version of that interaction effect: the output that emerges from the combination of inputs, not from the inputs themselves.
Equivalently, TFP is an error term. You build a model predicting how much output a given set of inputs should produce. TFP is the residual — the difference between prediction and reality. A high residual means the model missed important higher-order effects. A low or negative residual means the model was too optimistic about how well the inputs would work together.
The China HSR Example
China’s early high-speed rail rollout had enormous TFP. Connecting Beijing to Shanghai did not just move people faster — it combined two economies that were previously semi-isolated. The first-order effects (travel time savings, jobs created) were what the planners modeled. The higher-order effects (behavioral changes, new urbanization patterns, complementary logistics industries, network effects from connecting supply chains) were what the model missed. The residual was huge because the synergies were huge.
Later rollout to secondary cities had lower TFP. Not because the trains were worse or the labor less skilled, but because the synergistic potential was smaller. Connecting two mid-tier cities does not generate the same combinatorial explosion as connecting the two largest economic hubs. The inputs were similar. The interaction effects were weaker. The residual shrank.
This maps directly onto China’s economic trajectory: the investment-led model produced spectacular TFP when the infrastructure gap was enormous — almost any connection between previously isolated nodes created multiplicative value. But as the low-hanging synergies were harvested, each additional unit of investment produced less combinatorial magic. The model kept deploying capital at the same rate. The residual went negative. The bridges to nowhere are investments with negative TFP — the inputs go in, but the synergies do not exist, and the output is less than the sum of the parts.
TFP as Goodhart Victim
TFP suffers from a version of Goodhart’s Law. When governments target “productivity growth,” they optimize for the inputs they can control (capital deployment, labor mobilization) rather than the synergies they cannot. But TFP is the synergies. You cannot increase synergy by adding more inputs — that is the entire point of it being a residual. Pouring more concrete into a city that already has enough roads does not create network effects. It creates maintenance costs.
The Theory of Constraints sharpens this: once the binding constraint shifts from “not enough infrastructure” to “not enough demand for the infrastructure,” additional investment is improvement at the wrong bottleneck. The TFP residual is the system’s way of telling you where the constraint actually is — and a shrinking residual is the signal that you are optimizing the wrong thing.
Dimwit / Midwit / Better Take
The dimwit take is “TFP measures how good a country’s technology is.”
The midwit take is “TFP is just a residual — it measures our ignorance, not anything real.” This is technically correct but misses the point. Yes, with a perfect model that captured every higher-order interaction, TFP would be zero. But we do not have that model, and the size of the residual tells you something important about how much combinatorial value the system is producing beyond what linear thinking predicts.
The better take is that TFP is a legibility gap — the distance between what planners can see and what the system actually does. High TFP means the system is generating value through channels the planners did not design and cannot directly control. Low TFP means the planners’ models are accurate because the system has stopped surprising them. And a system that has stopped surprising its planners is a system that has stopped discovering new synergies — which is another way of saying it has stopped growing in the ways that matter.
Main Payoff
The practical value of thinking about TFP as synergy rather than technology: it reframes the question from “how do we get more productive?” to “where are the unexploited interactions?” Early-stage systems (developing economies, new platforms, young organizations) have high TFP because everything is connecting for the first time. Mature systems have low TFP because the obvious connections are already made. The way to raise TFP is not to add more of the same inputs — it is to find new nodes to connect, new combinations to try, new interactions the current model does not predict. The residual is where the interesting things happen.