Oskar Morgenstern, A New Look at Economic Time Series Analysis (1961)

Morgenstern’s 1961 study advocates renewing economic time-series analysis through spectral methods connected to econometric models. Moving from conventional business-cycle analysis to questions of stationarity and interdependence, it proposes investigating temporal structures rather than prescribing them through inherited classifications. Statistical measurement, however, cannot by itself provide economic explanation.

Morgenstern credits Wald with clarifying the limitations of conventional decomposition and treats Burns and Mitchell’s qualitative approach as sophisticated but at an impasse. Judgment-based selection of turning points and construction of averages may identify useful features without adequately distinguishing overlapping rhythms. His criticism nevertheless allows a role for the older approach:

This does not mean, of course, that this method may not yield insights into the behavior of time series which may be useful in economic analysis.

The alternative draws on communication engineering. Aggregate economic series can be approached as outputs of a system receiving both patterned inputs and chance disturbances. Daily routines, production schedules, agricultural seasons, and institutional calendars give substantive grounds for investigating periodic influences.

Consequently it ought to be possible to discover the regularities which are demonstrably present in many basic input series.

This inference qualifies the lesson economists drew from Slutzky: random inputs can produce apparently cyclical movements, but that possibility does not establish that actual periodic inputs become undetectable in aggregate outputs. Morgenstern invokes the ear’s ability to distinguish instruments within orchestral sound to suggest that aggregation need not render constituent rhythms inaccessible. The analogy motivates investigation without establishing which economic rhythms will be found.

Earlier applications of Fourier and periodogram analysis suffered from restricted evidence and computational limitations:

The efforts involved in the main only short time series and as a rule only a few coefficients were computed.

Electronic computation offers opportunities to overcome those limitations, provided it supports stronger analysis rather than merely accelerating existing routines. Spectral analysis measures the distribution of a series’ variance across frequencies. It need not assume perfectly recurring business cycles or confine the evidence to a predetermined decomposition. Its promise lies in distinguishing patterns obscured by aggregate descriptions.

Stationarity supplies a central qualification. Economic trends appear to conflict with the stationary processes assumed by spectral theory. Morgenstern considers removing trends before analysis and recognizing that an apparent movement in the mean may reflect strong low-frequency power within a stationary process. Changing variance and other forms of nonstationarity remain more difficult. Because an economic series supplies only one historical realization, statistical evidence cannot settle these questions independently of technological and institutional knowledge. The proposal remains a research program, not a declaration that economic data already satisfy its assumptions.

Discovery also precedes interpretation. An unfamiliar frequency should invite reconsideration of economic concepts and models rather than dismissal because existing theory does not explain it. Conversely, predictive success cannot alone establish theoretical truth. Spectral measurement is valuable insofar as it creates more demanding questions for explanation.

Morgenstern extends this argument from individual series to their relationships. Successive economic observations are dependent, complicating conventional correlation. Cospectral and quadrature spectral methods permit frequency-specific investigation of coherence and phase: relations between short-term movements may differ from those between long-term movements. A single correlation coefficient or overall lag can conceal these distinctions. His discussion of New York call-money and commercial-paper rates illustrates how longer-term components can exhibit greater lags than shorter-term components.

Applications to production scheduling, sales fluctuations, and stock-market analysis lead back to econometric modeling. Individual spectra provide only a beginning; relationships among series require models constrained by evidence and mathematical tractability. A persistent tension remains: longer records improve frequency analysis while increasing exposure to institutional and technological change. The study thus makes a disciplined case for finer temporal measurement that can challenge economic theory without confusing descriptive resolution or predictive performance with explanation.