Trend Analysis & Extrapolation

The Straight Line That Would Not Hold

Every strategy deck in the commercial space sector contains at least one line chart extrapolating a familiar trajectory: launch cost per kilogram declining, constellation deployments rising, broadband subscribers growing, debris population accumulating. The lines are drawn carefully, usually with a shaded confidence band, and they are almost always extrapolated forward as if the slope observed over the last decade would continue more or less unchanged into the next. The presentation is confident, the numbers are precise, and the conclusion — that whatever the trend currently supports will continue to support it — is rarely challenged by the audience.

The problem is that trends of this kind break in two characteristic ways, and extrapolation catches neither. They saturate, when the physical, economic, or political ceilings that bound the underlying system are reached and further improvement requires a different mechanism than the one that has produced progress so far. And they discontinue, when structural breaks — regulatory reversals, technological substitutions, geopolitical ruptures, demand shocks — invalidate the mechanisms that were producing the trend in the first place. A launch-cost curve that has fallen for a decade can plateau because the cost floor is approached; it can also break upward because a new regulatory regime raises operating costs faster than technology lowers them. A straight-line extrapolation misses both, and the strategies built on it tend to discover the failure the hard way.

Trend analysis and extrapolation, done well, is the discipline that takes the line chart seriously without being seduced by it. The method’s value is not in replacing extrapolation with forecasting but in structuring extrapolation so that its assumptions are visible, its ceilings are named, and its discontinuity risks are flagged before they become the thing that breaks the plan.

Statistical Forecasting, Systems Thinking, and the Moore’s Law Template

The method sits at the intersection of two traditions that mature practitioners use together. The first is the statistical forecasting tradition — linear and nonlinear regression, time-series decomposition, logistic growth models — that traces back through twentieth-century operations research and demographics. This tradition supplies the mathematical tools for characterizing trend shape, estimating rates of change, and producing confidence intervals around projections. It is rigorous on what can be measured and disciplined about what its measurements do and do not imply.

The second is the systems-thinking tradition associated with Jay Forrester’s system dynamics work at MIT in the 1960s and with the broader literature on feedback loops, reinforcing and balancing dynamics, and structural breaks. This tradition supplies the causal vocabulary that statistical forecasting by itself lacks. A statistical model can describe that two trends co-move; a systems-thinking model can explain why, identify whether the relationship is reinforcing or balancing, and reason about what would break the relationship. For anything beyond short-horizon extrapolation, the systems-thinking layer is what keeps the method honest.

Technology forecasting has become the most visible application of trend extrapolation. Moore’s Law — Gordon Moore’s 1965 observation that transistor density on integrated circuits was doubling every year or two — is the canonical case, and its fifty-year run made a generation of technology strategists comfortable with extrapolating exponential trends as if they were physical constants. Moore’s Law’s eventual slowdown (the approach to atomic-scale physical limits, the shift from transistor count to energy efficiency as the binding performance metric) is now the canonical cautionary tale: even the most durable-looking technology trends have ceilings, and the transition from “trend continues” to “trend saturates” is typically visible only in hindsight.

The space sector has its own Moore’s-Law-shaped stories — the decade-long decline in launch cost per kilogram, the S-curve acceleration of constellation deployments, the exponential-looking growth in debris objects in specific orbital bands. Practitioners who extrapolate these trends without the saturation and discontinuity disciplines are repeating the mistake the semiconductor industry made in the 2000s, and the correction, when it comes, is typically expensive.

What the Method Sees That the Line Chart Does Not

The characteristic analytical gesture of rigorous trend analysis is the systematic replacement of a single projected line with a structured set of trajectories, each carrying named driving conditions and flagged discontinuity risks.

Launch Cost, Constellation Growth, and the Feedback Loop That Follows

Consider the method applied to the well-known launch-cost trend for LEO access. Historical data traces a trajectory from roughly $50,000 per kilogram in the Space Shuttle era of the 1980s and 1990s, through an expendable-launch-dominated plateau in the 2000s at figures in the $10,000–$20,000 range, to approximately $2,700 per kilogram by the late 2010s and early 2020s with the widespread deployment of partially reusable medium-lift vehicles. The shape is not linear; it is an S-curve-shaped decline inflected by the introduction of reusability as a substrate technology, and the evidence for that interpretation is the step-change in slope that accompanied the reusability transition.

The drivers are identifiable. Commercial competition — multiple qualified providers pursuing launch-cost reduction as an explicit competitive strategy — accelerated the decline during the 2010s. Reuse rates per vehicle, climbing from zero to double-digit flight counts, produced per-launch amortization gains that compound over the fleet. Supplier ecosystem maturation reduced component and refurbishment costs. Each driver is robust on its own terms, but each has a ceiling.

The saturation analysis identifies the relevant ceilings. Propellant costs constitute a floor below which cost per kilogram cannot fall regardless of reuse count. Range and operations costs — insurance, ground infrastructure, regulatory compliance — impose their own floor. Reuse-cycle gains diminish as per-vehicle flight counts approach asymptotic limits. The baseline projection is therefore a plateau somewhere near $1,500 per kilogram by the early 2030s, as the current architecture approaches the envelope that its underlying technology supports.

The acceleration variant identifies the conditions for further reduction. A new class of vehicle architecture — fully reusable, higher-cadence, capable of amortizing fixed costs across flight rates an order of magnitude above current practice — could push costs below $500 per kilogram if the architecture achieves operational maturity. The conditions are specific: full reusability of both stages rather than first-stage-only, flight rates sustained at several times current levels, and a market capable of absorbing the volume without collapsing the economics.

The deceleration variant identifies the countervailing forces. A tightening of orbital traffic regulation — licensing fees, coordination requirements, debris-related compliance costs — could offset technology gains, producing a plateau above the baseline trajectory or even a partial reversal. An insurance-market repricing after a major debris event could produce a similar effect from a different direction. A supply-shock in specialized materials, propellants, or critical components could interrupt the decline entirely for several years.

The interaction matrix produces the most consequential finding. The launch-cost decline trend reinforces the constellation-deployment trend directly — lower per-kilogram cost enables larger constellations at the same capital outlay. The constellation-deployment trend amplifies the orbital-congestion trend. The orbital-congestion trend feeds back into the regulatory posture that constitutes one of the principal deceleration conditions for the launch-cost trend itself. The feedback loop is not speculative; it is visible in the data already. A strategy that extrapolates the launch-cost decline without recognizing this loop is extrapolating a trend against its own feedback structure.

The non-obvious insight is not in any individual trend line; it is in the loop. The industry is producing the conditions that will eventually slow the trend that is enabling the industry. Strategic posture depends on how fast the loop closes, which depends on how the governance response to congestion develops, which depends on signal detection that trend analysis alone cannot supply. This is the point at which the method hands off to scenario planning for the exploration of alternate trajectories and to horizon scanning for the weak signals that might foreshadow the discontinuity.

Where It Earns Its Keep and Where It Falls Short

The method’s strength is its empirical backbone. For foresight work that must be anchored in something other than speculation, trend analysis provides the observable trajectory against which scenarios can be defined as divergences. It is the natural first step in any foresight exercise because its outputs — shape, ceiling, drivers, interactions — define the strategic terrain that the subsequent methods operate on. Without it, scenarios float free of the data; with it, they are grounded in trajectories that can be challenged and revised as new observations arrive.

Its weaknesses are consistent with what extrapolation can and cannot do. The method assumes some continuity with the past; it fails at structural breaks and paradigm shifts by design, and its outputs should always carry the discontinuity flags that make this limitation explicit. Quantitative precision creates false confidence; a projection to three significant figures is not the same as an accurate forecast, and the method’s discipline is to present ranges rather than point estimates. Trend interactions are difficult to model rigorously without simulation tools, and the method’s qualitative interaction matrix is a useful simplification but cannot substitute for formal system dynamics in cases where the feedback structure is genuinely complex.

S-curve identification is clearer in hindsight than in real time. A trend that is on the steep portion of an S-curve is indistinguishable, on short data series, from a trend that is exponential indefinitely. Honest shape claims acknowledge this uncertainty and treat the S-curve identification as the most plausible reading given current evidence rather than as settled fact. Qualitative trends resist quantification; the method handles them with proxy indicators and milestone tracking, which is less satisfying than numeric precision but more honest than forcing numeric precision onto directional signals.

The method is not suitable as a standalone approach for highly uncertain, long-horizon topics. A ten-year extrapolation in a domain where the underlying mechanisms are themselves evolving is not a forecast; it is a baseline against which scenarios must be built. Scenario planning is the natural pairing, and trend analysis produces the baseline trajectory that scenario narratives diverge from. Horizon scanning covers the discontinuity flank that trend analysis cannot see from historical data alone. Three horizons analysis uses trend shape and saturation readings to calibrate horizon classification. Backcasting uses trend baselines as the counterfactual against which normative pathways are defined.

For the Practitioner