Food Crises in Germany, 1500–1871

5.1 The Trend of the Real Rye Price

Panel (a) of Figure 3 presents the three versions of the aggregate real rye price for Germany from about 1500 to 1871. All values are in natural logarithms. Panel (b) shows the trend component of the real rye price identified using the BW filter for all three samples over the same period. Panel (a) establishes that the price based on the smaller but geographically stable sample closely follows the overall movement of the larger but heterogeneous unbalanced sample that includes all 70 cities. Similarly, the price based on the geographically stable sample is mostly well in line with the one based on the stable sample of cities with 5 percent or less missing observations but shorter time coverage. This buttresses our confidence that the geographically stable sample represents a good compromise between data coverage and quality. Below, we therefore focus on the geographically stable sample.

Fig. 3

Three versions of the aggregate real rye price and corresponding trend components, Germany, 1512–1871. Note: Panel (a): natural logarithm of aggregate real rye price for three samples: unbalanced sample: 70 cities, varying data coverage; geographically stable sample: 9 cities from 2 geographically fixed areas (Northern and Southern Germany); stable sample: 9 cities with not more than 5 percent missing observations. Panel (b) Trend components from Butterworth filter (Pollock, Trend Estimation), with cut-off frequency 1/20 and order 4. Mean and trend from linear regression shown for geographically stable sample. Data sources: same as Fig. 2.

The trend component displayed in panel (b) indicates changes in the relative scarcity of grain. With a given level of technology, the mid-term movements of the trend component are driven by shifts in relative prices between land-intensive and labour-intensive products. This is because grain is a land-intensive product, whereas the CPI used for deflation also includes products with a considerable labour component, such as linen cloth and beer. Hence, in a Malthusian economy with little technological progress, the trend of the real rye price should parallel swings in population size: Population growth increases the scarcity of land relative to labour; hence, the price of land-intensive products should increase relative to labour-intensive products. By contrast, demographic contraction should produce a decline of the real price. ^[43]

In the sixteenth and the eighteenth centuries, German population increased by an annual rate of about 0.4 percent and declined by at least a third in the second quarter of the seventeenth century. In 1816–1870 the rate of demographic expansion doubled to 0.8 percent. ^[44] Following the hypothesis stated above, the real and relative prices of grain should fall in the era of the Thirty Years’ War and should increase during the remainder of the period under study. According to panel (b) of Figure 3, the mid-term movements in the trend of the real price of rye conforms to the expected pattern until the late seventeenth century: the real price of rye rose during the sixteenth century, declined between about 1630 and 1670 and recovered until the end of the seventeenth century. This pattern warrants characterizing Germany as a Malthusian economy during the first two centuries of the modern era.

The situation changed around 1700: Apart from two mid-term peaks in the 1760s and the 1790s and the dip during the agricultural price depression of the 1820s, the trend component of the real rye price remained essentially stable until the end of the period studied, which contradicts the Malthusian hypothesis. If anything, the trend was slightly downward after 1800 despite the acceleration of population growth mentioned above. A linear time trend fitted to the logarithm of the real price based on the geographically stable sample shows that it declined with an annual rate of 0.05 percent per year (p<0.05, Figure 3, panel (b), dot-dashed line). Also, after 1700 the mid-term peaks of the trend component do not reach the high levels prevailing during the sixteenth and seventeenth centuries.

Two factors account for the increased stability of the level of the trend component after 1700: the price of energy and increasing agricultural productivity. First, energy has a considerable weight in the consumer basket underlying the CPI – 11.0 percent in the 1760s, with firewood or charcoal alone making up for 5.5 percent ^[45] – and the relative price of firewood to rye began to increase continuously from about 1700. This reflects a rising energy intensity of economic growth related to the expansion of proto-industrial iron manufacture and other non-agricultural activities. ^[46]

Second, the textiles-rye price ratio remained more or less stable between about 1700 and the 1830s. ^[47] The most obvious explanation is that agricultural supply broadly kept pace with the expanding population. Estimates of grain production in two proto-industrial regions reaching back to the 1730s suggest that this was indeed the case. ^[48] One potential reason for the expansion of grain output was the advance in grain market integration during the late seventeenth and early eighteenth centuries. ^[49] The resulting reduction of trade costs must have improved the efficiency of the allocation of factors of production. The behaviour of the real rye price thus establishes that Malthusian forces began to lose sway over Germany from about 1700. This conclusion departs from earlier research by Wilhelm Abel who posited that Malthusian forces continued to prevail until the middle of the nineteenth century. ^[50]

5.2 A Chronology of Food Price Crises

Panel (a) of Figure 4 again shows the logarithm of the real rye price for the geographically stable sample from 1543 to 1871. Panel (b) shows the cyclical component of the real rye price both for the geographically stable sample and the sample with a fixed composition of cities (the stable sample). Our discussion again focuses on the results for the geographically stable sample and uses the information on the stable sample as a cross-check. The peaks of the cyclical component of the geographically stable sample (marked with vertical bars and corresponding years at the bottom of panel (a) in Figure 4) indicate food price crises (when the cyclical component of the real price exceeds the trend by at least 20 percent). We regard contiguous years with high prices as one single crisis.

Fig. 4

Aggregate real rye price and food price crises in Germany, 1543–1871. Note: Panel (a) plots the natural logarithm of the aggregate real rye price for the geographically stable sample. Vertical lines and years indicated on the horizontal axis in panel (a) mark major price peaks (cyclical component exceeds trend component by at least 20 percent). The grey upper horizontal axis in panel (b) and corresponding crisis years refers to our robustness check using the rye price based on the stable sample. Data sources: same as in Figure 2.

We find 24 price crises within the period studied based on the geographically stable sample with a total of 44 crisis years; 15 of these crises are multiyear crises. Table 2 presents an overview of all food price crises in chronological order. Columns 2 and 3 show the years in which the individual crisis began and ended, column 4 shows the number of corresponding crisis years and column 5 the year within the crisis period in which the rye price reached a peak. Columns 6 to 8 show the percentage deviation from the trend in the pre-crisis year, the first crisis year and in the peak year, respectively. In addition, column 9 presents the cumulative percentage deviation from the trend over the whole crisis period including the pre-crisis year. In column 10 we rank all crises based on this criterion. To classify the crises as dearth, famine or subsistence crises, we also include demographic data if available (columns 11 to 14).

While large and long-lasting deviations from the trend are informative about the level of economic stress, we would like to know more about these crises: were they part of European-wide shocks? What do we know about their demographic impact? Hence, we contextualize all 24 crises using information from the current historiography on local, regional, national, and European crises. To this end, we use the overview by Guido Alfani and Cormac Ó’ Gráda, who develop a European chronology of famines based on the criteria of high grain prices and high mortality in which at least two of nine macro-regions of the continent were affected. ^[51] The chapter by Dominik Collet and Daniel Krämer on Germany, Switzerland and Austria in the same volume serves as our benchmark at the country level. ^[52] The quantitative approach of our study complements these overviews in two ways. First, we date and rank crises within a unified methodological framework and discover hitherto unknown food price crises. Second, we draw on vital rates from recent historical demographic work to check whether food price crises also qualify as subsistence crises. ^[53]

While Collet and Krämer also discuss evidence on famines dating back until the fourteenth century, ^[54] our crisis chronology starts only in the middle of the sixteenth century due to data limitations. Visual evidence from the unbalanced sample in Figure 3 in Section 5.1 suggests that there certainly was a major food crisis in the first half of the 1530s. ^[55] Figure 3 also suggests a likely strong earlier shock in 1517. ^[56] We do not include these two crises in our chronology, because the geographically stable sample does not cover these years.

Whenever possible, we complement our discussion of individual crises with information on local and regional events in Germany, which is mostly based on local prices as well as demographic data where available. In a few cases, output data (tithe or yields) are available as an additional indicator of yield quantity; low yield quality is traced with information on ergot poisoning, medical descriptions of which surfaced in the late sixteenth century. ^[57] Similarly, we list three crises that do not feature in our baseline result but are mentioned in the existing literature.

The first two food crises we identify when using our methodology occurred in 1546 and 1551/1552. Neither of them has been documented in the existing literature, but the second price crisis coincides with a marked downturn of tithe returns, which proxy grain output, around Nuremberg. ^[58] The first major famine in our chronology that is covered by the existing literature is the one of the early 1570s. ^[59] The cyclical component of the real rye price in 1572 is relatively small, but note that the trend component experienced a massive rise around 1570 (Figure 3). During several years of the 1570s, the trend component as such is more than 20 percent above the long-term mean shown in Figure 3, panel (b). There were only three such episodes in the entire period of observation. ^[60] Thus, our analysis suggests that the severity of this crisis, which is attested by qualitative accounts, resulted from the combination of forces operating in the medium term and short-term shocks. The price crisis of 1587 was a European crisis but has not been documented for Germany yet. There is regional evidence for an ergot epidemic in Silesia in 1587, and there was considerable excess mortality in Nuremberg, albeit slightly earlier in 1585/1586. Evidence from both regions is consistent with low food availability around this time. ^[61]

The end of the sixteenth century most likely witnessed a low level of food security. The 1590s saw the second episode in which the trend component was 20 percent above the long-term mean over several years. The real rye price rose almost continuously, culminating in a peak in 1599, which falls only slightly short of the criterion of 20 percent excess above trend (19.6 percent), and is thus not included in Table 2. In our alternative chronology based on the CF filter, the deviation is slightly larger than 20 percent, but when using the sample with a fixed composition of cities, 1599 is again just below the threshold (see below, section 5.3, and Figures 4 and B1). For 1595/1596 ergot epidemics are documented in eight regions, which suggests a severe famine. Low food availability at the end of the sixteenth century is also rendered plausible by below average yields in Franconia, and moderate excess mortality recorded for Nuremberg in 1599/1600. ^[62]

The price spike of 1612 does not appear in existing chronologies of food crises. Nevertheless, narrative evidence for Ulm (northern Swabia) documents consecutive years with cold winters during the first half of the 1610s, notably in 1611/1612, leading to a shortage in grain. In 1611 and 1614 public authorities distributed grain at a reduced price to the inhabitants of the town. The food crisis followed a plague epidemic peaking in 1611, which may have disrupted the production and distribution of food more severely than it reduced demand. ^[63] The crises of 1625/1626 and 1634/1635 have been described on the basis of local data on prices, tithes, and mortality for Nuremberg and central Franconia (specifically for 1625–1628, and 1632–1635). Both crises again coincided with massive plague epidemics, which may have adversely affected food supply. The food crisis in the 1620s has been noted at the European level, whereas the one in the 1630s may have been confined to Germany. ^[64]

Existing chronologies also mention a food crisis in 1620–1623, which does not feature in our results. ^[65] While European annual, summer, and winter temperatures were far below average at this time (mean -1.5 standard deviations, SD), these years do not belong to the coldest recorded (mean -2 SD). ^[66] This was, however, also a period of severe currency debasement related to inflationary war financing during the initial stage of the Thirty Years’ War, the so-called Kipper und Wipper era. Whereas nominal prices of all products quoted in debased currency soared during the half decade preceding 1623, the real price of rye was actually below trend around 1620 (see Figure 4, panel (b)). This may reflect an effect of the inflation tax that shifted purchasing power from workers, whose nominal wages remained sticky, to warring rulers and thereby reduced demand for food. Thus, the Kipper und Wipper era likely was a serious economic crisis, but it was not triggered by an adverse supply shock in agriculture that is characteristic of a food crisis. ^[67]

The crises of 1649-1652 and 1661-1663 have been noted by several studies and are ranked third and second in severity among all 24 crises (column 10 in Table 2). ^[68] Both crises show cumulative sums of deviation from trend of more than 140 percent and stretch over three and four years, respectively. Regional evidence confirms the severity of these food crises. Specifically, widespread ergot poisoning was reported for the Vogtland (Saxony) in 1648/1649. ^[69] In Brandenburg and its western possessions, a poor harvest led to a ban on grain exports in 1651, and similar measures were taken in 1660–1662. ^[70]

The crisis of 1674–1676 was of similar magnitude according to our ranking criterion (see column 10 in Table 2). Its severity is confirmed by evidence from demographic data, which is available from 1675 onwards. We show a cumulative deviation from trend established using the BW filter of the logarithm of the series (specification and calculation of percentage deviation as for the real rye price) during periods of crisis and the first post-crisis year (in columns 11 to 14 in Table 2). Research suggests that vital events reacted both immediately and with a time lag on fluctuations of material living conditions. ^[71] Thus, we measure the demographic impact of a food crisis with the sum of the deviation from trend in crisis years and one year later. During the crisis of 1674–1676, a sample of parishes located in Southern Germany recorded exceptionally high numbers of deaths and a dip in the number of births. There was also renewed ergot poisoning in Vogtland and Westphalia in 1675. ^[72] Thus, the grain price spike of 1674–76 clearly qualifies as a subsistence crisis.

The crisis of 1685 features a comparably small deviation from trend (just above 20 percent) and lasted only one year. Evidence concerning vital events in Southern Germany is not consistent with a subsistence crisis. Consequently, 1685 may be best characterized as a dearth episode. By contrast, the crisis of the early 1690s is known as one of the most severe famines in early modern Europe. It is clearly visible in a spike of the number of deaths and a low number of births in a sample of southern German parishes. ^[73] The crisis of 1699/1700 has been described for Brandenburg-Prussia, North-Western Germany and Saxony. There is also testimony of widespread ergot poisoning in Thuringia in 1699. Given the massive spike in the aggregate rye price (rank 5 according to column 10 of Table 2), this crisis may have been more severe than the crisis of 1693/94 according to the cumulative deviation from trend. While it is not visible in the vital events series based on parishes in Southern Germany, a tentative estimate of vital rates at the aggregate level suggests a spike in the death rate and a trough in the birth rate in 1699, which clearly characterizes this episode as a subsistence crisis. ^[74]

The food crisis of 1709 has been documented using qualitative evidence for Brandenburg-Prussia, and its demographic effects are reflected in a high number of deaths in 1709/1710 and a low number of births in 1710 in Southern Germany. It has also been characterized as a European-wide crisis but its impact on mortality was weaker compared to the crisis of the early 1690s in most countries. ^[75]

Dominik Collet and Daniel Krämer mention a food crisis in 1724/1725 in their chronology. We find spikes in the real rye price in these years, but the deviations from trend are less than 20 percent (8.9 percent and 12.7 percent, respectively). In a tentative estimate of aggregate vital rates for this period, there is also a minor spike of the death rate and a small dip in the birth rate in 1724, which combine into a negative rate of natural increase. Finally, contemporary reports mention widespread ergot poisoning in Silesia, parts of Pomerania and Brandenburg in 1722/1723. All this renders it plausible that at least a minor subsistence crisis occurred around 1724. ^[76]

The crises of 1740 and 1771/1772 are well-known pan-European food crises. However, the one of the early 1770s was apparently much more severe in Germany than in other parts of the continent. ^[77] The greater severity of the crisis in 1771/1772 relative to the one of the early 1740s is confirmed by our ranking according to the cumulative deviation from trend (rank 6 vs. 8). The crisis of the early 1770s was also characterized by the highest spike in the crude death rate in the aggregate series of life events starting in 1730, and a marked slump in the crude birth rate, rendering it one of the most serious subsistence crises in German demographic history. By contrast, the price peak in 1757 is less known. Nevertheless, this crisis can be clearly identified by an excess of deaths over births in the aggregate series of vital events.

The price spike in 1795 is the last food supply shock that went together with a massive increase of mortality, which – combined with a slump in fertility – produced an excess of deaths over births at the aggregate level. The one of 1802, by contrast, seems to have been a minor crisis; to our knowledge it is mirrored neither in demographic evidence nor in contemporary reports.

From the 1810s to the 1850s, four negative shocks in food supply occurred at short intervals: (i) The first was the crisis of 1816-1818 following the eruption of Mount Tambora in 1815, which resulted in crop failures over consecutive years. ^[78] This crisis was the most serious one in the period under study with regard to the sum of price deviations from trend (column 10 in Table 2). (ii) The substantial spike in the real price of rye in 1831/1832 does not appear in existing famine chronologies, but deviations of vital events from trend are consistent with a famine, which is corroborated by qualitative evidence for parts of Prussia. ^[79] (iii) Serious food shortages in 1846/1847 originally arose from the European-wide potato blight spreading from 1845, the effects of which were aggravated by crop failures of cereals in 1846. ^[80] (iv) The price spike of 1854 was less marked than the one in 1846/1847, which makes this crisis less known, but the mid-term trend of the real rye price shows a massive bulge during the first half of the 1850s (see Figure 3). The economic effects of the combination of a high trend level and a short-term shock were likely of similar severity as those of the shock of 1846/1847: In 1855, the birth and marriage rates showed their lowest values in the entire period between 1815 and 1913, and emigration to overseas destinations reached a peak in 1854. There were also reports on widespread ergot poisoning in several regions. Net industrial capital formation was zero in 1855 and strongly negative in 1856, which suggests that the crisis also affected the nonagricultural sectors. ^[81]

The crisis of 1868 was the last economic crisis that originated from output shortfalls in the agrarian sector. While emigration to overseas destinations shows a local peak in 1868, the effect on vital rates was small at best. The only exception was in the extreme northeast of the country, the Province of East Prussia, which experienced an excess of deaths over births in 1868 and where there was a high number of deaths attributable to ergot poisoning. ^[82]

The food crises from the late 1810s to the 1860s share the characteristic that their demographic effects were weaker compared to the crises of the eighteenth century. Notably, the cumulative deviation of the death rate from trend never exceeded 10.3 percent, whereas the deviation was always larger than 23 percent in the food crises that occurred between 1740 and 1795. As a result of the weakening of the response of the death rate to food price shocks, there was no year with a negative rate of natural increase – that is, an excess of deaths over births – at the aggregate level after 1815. The last year that saw a negative rate of natural increase was 1814. It may have been related to a spike of grain prices in 1812 (cyclical component 12.2 percent), which, however, does not qualify as a food crisis according to our criterion. ^[83] A weakening of the demographic impact of dearth is also documented at the European level, in particular with respect to the Tambora crisis. ^[84] Nevertheless, high food prices were still accompanied by a decline in the birth rate (column 12 of Table 2) and, hence, the rate of natural increase, a decline in the marriage rate and, increasingly so, spikes in emigration. There were also substantial regional differences. Regions in north-eastern Germany, where the lower classes depended mostly on employment in agriculture, continued to suffer negative rates of natural increase during the crises of 1831/1832, 1846/1847, 1854, and 1868. ^[85]

5.3 Robustness Checks

We performed four robustness checks to explore the sensitivity of the chronology of the 24 identified food crises and their relative ranking with regard to sample variation and econometric filter. We derived an alternative chronology using the same BW filter but based on the stable sample (as described in Section 3, see plot of cyclical component and corresponding crisis years on grey axis at the top of panel (b) of Figure 4 in Section 5.2). In this alternative version, only 19 crises are detected. Three crises cannot be detected simply because of the shorter period of observation: 1546, 1551 and 1867. Thus, 21 of the 24 crises are left, which are detectable in the stable sample. Three crises do not pass the threshold value and an additional one is detected. The three crises that are missing when using this alternative sample are the ones of 1612, 1634/1635, and 1795. The crisis of 1612 is ranked 17 of 24 in the baseline results; 1634/1635 is a multi-year crisis and has rank 14. In the stable sample the cyclical components for these years are below the threshold value (18.9 percent and 16.3 percent) but still considerably large. In the baseline specification, the deviation from trend in 1795 is just above the threshold of 20 percent and ranked 23; the value based on the stable sample fails to meet this criterion only slightly (19.9 percent). The additional crisis detected with this sample is 1805, lasting one year with a peak of 23.8 percent; the deviation in the geographically stable sample is only 13.8 percent. 1805 was marked by a negative deviation of average temperature, ^[86] but the demographic response to the crisis was weak and ambiguous. While fertility at the aggregate level deviated negatively from trend in 1805-1806 (-2.8 percent), the death rate also decreased (-4.7 percent). Thus, there was no excess mortality, which contradicts the typical pattern during famines.

The ranking of the top half of the 24 crises changes only slightly (see Table B1 in the Appendix). The most notable change is that the crises of 1685 and 1854 are ranked higher (12 vs. 21 in baseline and 11 vs. 18 in baseline, respectively); by contrast, the one of 1693 has a lower rank (14 vs. 9 in baseline).

Our crisis chronology is also robust when altering the method of time series decomposition to the CF filter. Using this filter, only 23 crises are identified; we show them in the appendix (see Figure B1 and Table B1, Appendix B). The crises that are missing compared to the results based on the BW filter are those in 1546 (at the beginning/end of the sample the CF filter might be unreliable), ^[87] and 1795 (which is just below the threshold of 20 percent). The year 1599 is an additional crisis year (just above the threshold, see Section 5.2 for a discussion of the historical evidence). There are two minor changes regarding the dating of crises: 1634/1635 is reduced to a single-year crisis (1635), and the crisis of 1674–1676 starts one year later in 1675. The ranking of crises is very similar (see Table B1, Appendix B). Among the top three crises, 1816 and 1661 swap positions. Among the less severe crises, 1551 is now ranked 13 (vs. 12 in baseline) while 1709 is attributed rank 12 (vs. 16 in baseline).

Next to varying the sample and the type of filter, we considered two alternative cut-off periods for the BW filter, P=30, and 60, respectively. Intuitively, a shorter cut-off period attributes more fluctuations to the trend component whereas a longer cut-off period generates a smoother trend (see also Figures B2 and B3, Appendix B). When there are runs of years with high prices, a smoother trend would make it easier to identify a multi-year crisis, because the decomposition attributes less variation to the trend and more to the cyclical component.

The majority of crises is insensitive to these variations; nevertheless, the following modifications appear: The detection of crisis years at the sample end points is different (1540s, 1860s), but in a rather unsystematic way. With a higher cut-off period (P=30, P=60), 1685 (ranked 21 in the baseline) is not a crisis year anymore. More importantly, the crisis of the early 1570s grows in significance as specifications with longer cut-off periods show a higher number of crisis years. In the specification with P=60, 1571–1574 is ranked at 6. In addition, the year 1599 is detected. These results are in line with the strongly rising trend component of the baseline results discussed in Section 5.2 above; with a higher cut-off period more of the variation is attributed to the cyclical component.

These alternative results also make it clear that the importance of the absolute price level is downplayed by trend-cycle decomposition in general. A smoother trend at a longer cut-off period shows a lower level during the 1570s and 1590s, so that the shock (the cyclical component) grows in size, which partly remedies this issue. However, with P=30 the trend level is underestimated at the beginning of the sample when compared to the long-run sample, giving an implausibly high weight to the crisis of the 1540s (see Figures 3 and B2). Furthermore, at higher cut-off periods such as P=60, it also becomes more difficult to argue that the cyclical component represents unexpected shocks to which adaptation is more difficult than to the current trend.

We conclude that our chronology of food crises is largely robust to both sample variation and the method applied to identify cycles in the aggregate real rye price.

6.1 Population Growth and Climate Change (c. 1540–1645): Rising Demand Meets Increasingly Constrained Supply

Food crises do not seem to have evolved in any systematic pattern during the sixteenth century with respect to frequency, duration and deviation from trend. However, the trend component of the real rye price experienced a massive rise over time (see Figure 3 and the previous discussion in Section 5): From about 20 percent below the long-term average at the beginning of the century it rose to a level slightly below the mean during the 1540s and 1550s, and peaked at values more than 20 percent above the long-term average in the early 1570s and the 1590s. Thus, whereas the local peaks of the real rye price in 1572, 1587 and 1599 deviated less from trend than the majority of other food crises, the absolute value of the real rye price in those years was at par with the value prevailing in 1642–1652 and 1661–1663, which were years with some of the most serious food price crises according to our ranking. In other words, the food crises of the last three decades of the sixteenth century represent particularly dire episodes in a context where access to food was severely restricted due to very high price levels.

Explanations of the situation in the sixteenth century can be structured by considering the aggregate supply and demand for grain. The first explanation relates to an important supply side factor: climate change. From about 1560 the weather in Central Europe became wetter and colder in all seasons. This increased the likelihood of harvest failures and, correspondingly, of dearth and subsistence crises. The negative relationship between grain prices and temperature implies a structural deterioration in food availability during this period. ^[90]

The second explanation of the changing behaviour of the real rye price during the sixteenth century relates to the demand side. Between 1500 and 1600, Germany’s population grew by 78 percent from about 7.2 million to about 12.8 million. In line with the textbook model of a Malthusian economy with static technology and a fixed amount of land, the real wage of urban labourers fell by roughly one-half, because population growth reduced the endowment of labour with land. The mid-1590s recorded the lowest level of real wages during the early modern era. Thus, while the real price of rye rose to an extraordinarily high level, income per capita declined; in combination, these two processes reduced the quantity of food that households could demand. ^[91] Note that population growth may also impact on the supply side: While the amount of land is fixed, not all plots are of the same quality and increasing reliance on marginal land in the wake of the expansion of the cultivated area may have decreased the level of land productivity and increased the variability of harvest outcomes, given that marginal land is more susceptible to shocks. ^[92]

6.2 The Maunder Minimum (c. 1645–1715): High Frequency and Intensity of Food Price Shocks

The Maunder Minimum (MM) characterizes a climatic period with low solar irradiance, which contributed to low temperatures during the growing season and extremely cold winters. ^[93] The volatility of the real rye price, measured by the standard deviation of the cyclical component, was 21.7 percent during this period, compared to only 13.6 percent during the preceding phase. The 70 neighbouring years (35 years before and after the MM) feature a volatility of only 12.6 percent. Thus, dearth episodes indicated by rye price spikes were both more frequent and more severe during the MM than before and after. According to Table 2, five of the twelve most severe price crises of the whole modern period occurred during the MM. In addition, the MM exhibits an excess in crisis years: With a total of seven dearth episodes (only 1685 and 1709/1710 do not figure among the twelve most severe crises), this 70-year period contains seventeen crisis years, while the 70 neighbouring years include four dearth episodes with only seven crisis years. Finally, the total cumulative percentage deviation above trend was 638 percent during the MM period, but only 226 percent during the 70 neighbouring years. The demographic evidence that increasingly becomes available from the second half of the seventeenth century qualifies the dearth episodes of this period as subsistence crises except for the crisis of 1685. For this latter case, reports concerning ergot poisoning and export bans provide qualitative evidence of severe restrictions on access to food. The Maunder Minimum period implied that the preceding crises of the 1620s and 1630s, which were aggravated by plague epidemics, were followed by another eighty years of frequent and massive adverse shocks. The mid-seventeenth to early eighteenth century thus presents itself as a prime case of the economic effects of climate change in German economic history.

6.3 The Late Malthusian Disequilibrium (c. 1715 to early 1810s): High Vulnerability to Fewer Shocks

The century preceding 1815 has been characterized as a demographic disequilibrium because of a parallel decline of the death rate and the real wage; there was thus no long-term equilibrium relationship between the real wage and mortality as implied by the Malthusian hypothesis of a positive check. The period also qualifies as late Malthusian because there was still a negative relationship between population and the real wage, whereas other elements of a Malthusian economy had vanished. Recall the early signs of an increase in total factor productivity in agriculture mentioned in Section 5.1. ^[94] According to the evidence presented in Table 2, rye price crises were less frequent and less severe during this era than both before and after, and price volatility was relatively low (12.2 percent in 1716–1809). However, a salient feature of this period is that while the level of mortality fell, the death rate reacted much more strongly to fluctuations of the real wage than in other European countries at the time. ^[95] These fluctuations were essentially driven by grain prices. The German population was thus highly vulnerable to adverse shocks in agricultural output. Whereas the frequency and severity of food shortages declined after the end of the Maunder Minimum, they continued to have dire economic consequences and caused immense human suffering; the food crisis of 1771/1772 is a prime example.

6.4 The Post-Malthusian Era (late 1810s to 1860s): High Food Price Volatility Combined with Decreasing Demographic Impact

The period between the late 1810s and the 1860s qualifies as a post-Malthusian era for two reasons: First, the real wage remained stable and there was a progressive increase of the growth rate of real GDP per capita, despite a massive acceleration of population growth. The Malthusian effect of population growth that would dilute the resource endowment of labour was thus fully compensated by forces that raised total factor productivity. Second, as already mentioned above, the death rate was largely insensitive to negative income shocks at the aggregate level; by the late 1810s, the Malthusian positive check had disappeared in a short-term perspective as well. ^[96] However, during the first part of the post-Malthusian era, rye price volatility surged to 21.2 percent for the years 1810–1859 (19.7 percent for the years 1810-1871), that is, approximately the same level as during the Maunder Minimum period. In less than fifty years, there were four dearth episodes, three of which ranked among the top half of all food crises with respect to our severity indicator (see Tables 2 and B1).

Two explanations stand out to account for the increase in the frequency and magnitude of food price spikes between the late 1810s and the 1860s. The first explanation pertains to climate. Apart from the early 1820s, when weather conditions favoured grain farming (see the negative spike of real rye prices in 1824 in Figure 4), most of the period was characterized by a high frequency of negative temperature anomalies. Conversely, the warming that took place in the 1860s may have contributed to the decline of food price volatility in combination with globalization forces. ^[97]

A second explanation points to the possibility that third market effects increased grain price volatility between the late 1820s and the onset of the so-called European grain invasion in the 1860s. ^[98] From the 1760s, the northern parts of Germany had gradually emerged as an important supplier of grain to industrializing Britain. The enactment of the Corn Laws in 1815 temporarily muted this trend, but their progressive dismantling between 1828 and 1846 led to a renewed increase of grain trade between Germany and Britain and to a narrowing of the price gap between English and Prussian grain markets. The rise of the trend component of the real rye price between the 1820s and the 1850s thus reflects an increase in both national and international demand. However, trade volumes were highly volatile: peaks were recorded in 1818, 1830, 1841 and in the late 1840s. ^[99] Three of these peak years correspond with the food crisis years in our rye price series. This correspondence between food price crises and temporary peaks of grain exports is consistent with views originally presented by George Grantham and Amartya Sen.

Grantham, based on the work of Jean Meuvret, has argued that in a first stage of market integration, when trade costs were still high, long-distance trade of grain was intermittent and largely confined to periods of strong spatially asymmetric shocks. Only when prices were very high in one place and low in another were price gaps large enough to cover trade costs. ^[100] Sen highlighted the role of demand failure in food crises. ^[101] In regions where agriculture was the prime source of employment, harvest failures also led to a decline of labour income among the lower class because fewer harvesters and threshers were needed. Because of the decline of local demand, more grain became available for being exported to regions where households depended less on agricultural output for their income. At least at a first glance, the views of Grantham and Sen are consistent with the pattern of food crises during the first part of the post-Malthusian era, that is, high volatilities of grain prices and grain exports as well as the concentration of crisis-related mortality peaks in the north-eastern parts of the Kingdom of Prussia, which had specialized in export-oriented grain production. ^[102]

However, the question remains: if Prussian export regions were suffering from serious food shortages, why did high food prices no longer lead to a massive increase of mortality at the aggregate level between the late 1810s and 1860s compared to the eighteenth century? Three factors were likely responsible for this phenomenon. First, in contrast to the international level, market integration within Germany may have reduced the occurrence of regional deficiencies in supply much better than before 1815. Market integration was promoted by territorial consolidation and trade reforms during the Napoleonic Wars, the formation of the customs union (Zollverein) in 1834, and infrastructure development like paved roads construction (from the late 1810s) and railways (from the 1840s). ^[103] Second, the spread of potato cultivation reduced food risk. Potatoes react differently to weather conditions than do cereals, so harvests of rye and potatoes are only moderately correlated. Consequently, the spread of potato cultivation complemented market integration with regard to diversifying sources of food supply. ^[104] Finally, increasing state capacity and rising per capita income may have increased the volume and efficiency of food aid programmes implemented by both public authorities and private associations. ^[105]

In Germany, food crises ended in the late 1860s. This end of the historical pattern of food crises in the late 1860s is testified by the complete absence of a mortality response to the food price shock in 1867. Two reasons most likely account for the heightened resilience with respect to adverse shocks in agricultural output around this time. First, the progress of industrialization decreased the dependence of household incomes on agricultural output, which stabilized domestic food demand. Household survey data for industrial workers from 1888/1890 show how industrial incomes contributed to cover the costs of the required nutritional intake. However, German workers still had a low nutritional intake and faced a nutritional gap compared to the USA; consequently, significant welfare gains continued to make emigration attractive. ^[106]

Second, and notwithstanding rising agricultural protectionism from the 1880s, the globalization of grain markets allowed meeting the food demand of industrial workers through international markets. This is evidenced by the fact that Germany, after having been a net exporter of grain for many years (including the crisis years of the mid-1850s), showed a surplus of grain imports over grain exports in 1867 and 1868; over the following decade, Germany turned from an exporter to an importer of grain around 1870. ^[107]

Choice of Filter

Earlier research using grain prices to detect food crises applied the MA and the HP filter (cf. section 4). However, both filters have serious disadvantages. The trend estimate from a symmetric MA filter is by construction shorter at both ends of the sample, because leads or lags of the underlying time series are not available. Use of the HP filter has been discouraged by Hamilton. ^[108]

The CF filter used in the literature on commodity prices is a bandpass filter, that is, it targets a range of frequencies, which pass the filter while the other ones do not. This filter can also be used as a high-pass filter if the minimum period is set to 2. In its baseline version, the CF filter assumes series that are integrated of order 1; nevertheless, it can also be applied to stationary time series. ^[109] The price series we analyse turn out to be stationary or trend-stationary depending on the specification of the price series and the test chosen (Table A1).

The BW filter developed by Stephen Pollock can be implemented as a low-pass or as a high-pass filter (only low or high frequencies pass the filter); it is applicable to both stationary series and series with a unit root. ^[110] Thus, both filters can be used to achieve our aim of isolating the cyclical component as an indicator of food crises. Compared to the CF filter, the additional advantage of the BW filter is that it does not alter the frequencies in the passband (the gain of the filter is unity) except around the cut-off frequency. ^[111]

Specification of the BW Filter

The choice of a periodicity of 20 years is justified on two grounds: previous research and the periodogram of the detrended data. In view of potential longrun cycles, the literature on commodity prices decomposes the latter in three components using the CF filter: a long-run trend with periodicities larger than 70 years, medium-run cycles with a duration of 20 to 70 years, and short-run cycles below 20 years. ^[112] Medium-run cycles are often driven by demand and are followed by a supply response. In our context, demand for rye would increase with demographic expansion, and the supply response could result from increased specialization (as a result of market integration or learning-by-doing) and/or land expansion. These arguments also answer the question of what drives the long-run trend; the difference between long-run trend and mediumrun cycle is mostly about how long it takes for these structural adjustment processes to unfold. In view of food crises, our aim is to identify the short-run component, not the medium-run cycles.

Furthermore, much lower cut-offs, for example, in the range of three to five years such as used for the analysis of modern business cycles, are at odds with empirical results on pre-industrial grain prices. Spectral analysis of historical rye price series for Germany found cycles with a length of up to 13 and 10 years, respectively. An investigation into European grain prices has established periodicities of 5 and 16 years. ^[113] The associated price fluctuations should be part of the cyclical component and justify the choice of a longer cut-off periodicity.

In addition, as advised by Pollock, we inspected the periodogram to find frequencies that occur seldom in the data. ^[114] These frequencies form a natural criterion for the choice of the cut-off frequency of the filter. The periodogram of the linearly detrended natural logarithm of the real price for our preferred sample, the geographically stable sample, exhibits low spectral density between approximately 8 and 21 degrees, see Appendix Figure A1. These values correspond to a period of P=360°/8°=45 years (frequency f=1/45) and P=360°/21°=17.14 years (frequency f=1/17.14). Our cut-off frequency of f=1/20 corresponds to 360°/20=18°. ^[115]

In our robustness checks, we consider two alternative cut-offs. The cut-offs can be justified with another lower bound of the medium-run cycle of commodity prices, which is used less often in the literature (P=30). In addition, P=60 can be regarded as the upper bound of the medium-run cycle that falls in between the values of 50 and 70 years, which are mentioned in the literature. ^[116]

Next to the cut-off frequency, the filter requires the choice of the order, which we set to 4 in our baseline specification. A higher order potentially improves the transition from the passband to the stopband of the filter (cleaner separation of frequencies); however, it increases the complexity of the filter and comes at the cost of potential numerical instability. If the cut-off frequency is lower than the mid-point of 1/2 π (as in our case: the angular frequency corresponding to f =1/20 is ω =2 π f =2 π 1/20 = 1/10 π), higher values of the filter’s order become less important and lower orders suffice to separate the frequencies. ^[117] In our robustness checks, we also keep the order of the filter at 4 for the cut-o ff frequency of 1/30 but reduce it to 2 for f=1/60.

Fig. A1

Periodogram of linearly detrended logarithm of aggregate real rye price, 1543–1871. Note: real rye price based on geographically stable sample.

Specification of CF Filter in Robustness Check

To implement the CF filter, we linearly detrend log(R_t) and use the residual, because the filter cannot be applied to trending series (cf. the trend shown in panel (b) Figure 3). To use the CF filter as a high-pass filter, we specify the minimum period as 2 and the maximum period as 20. ^[118]