Feeding the Capital: Hyperlane Advisory Closures and Source-World Harvests as Sources of Variation in Grain and Protein Arrivals and Wholesale Prices at Coruscant, 30–24 BBY
Abstract
Coruscant is treated here as a capital that depends on food carried to it by hyperlane from agricultural worlds, a premise we adopt and do not measure. We ask how much of the variation in quarterly arrivals of grain and protein, and in their wholesale prices, is explained by advisory closures of the supply routes and how much by harvests at the source worlds. The records are the quarterly returns of the Coruscant provisioning registry for six routes over 25 standard quarters, 30 to 24 BBY, joined to Republic Hyperspace Survey Office advisory bulletins and to harvest returns lodged by the source worlds (147 usable route-quarters, 50 commodity-quarters). On the scale of the natural logarithm, and averaging the two orders in which closures and harvests can enter after season and trend, closures accounted for 38.5% of within-route variance in arrivals (bootstrap 95% interval 22.6–51.2) and harvests for 12.5% (5.6–21.0). For wholesale prices the ranking reversed: harvests accounted for 33.4% (13.9–44.9) and closures for 2.9% (0.5–11.4). A further ten points of a quarter closed went with 15% lower arrivals on the route. We think the ranking is likely to hold, though it rests on six routes, on harvest returns the source worlds report themselves, and on shares that depend on the order of entry. Stock and diversion records are needed to say why closures move arrivals far more than prices.
1. Introduction
In the fourth standard quarter of 29 BBY, advisory closures kept the fourth of the six supply routes studied here shut to civil traffic for 31.0% of the quarter's hours, the longest closure recorded on any route in the study. Arrivals on that route that quarter totalled 8.7 thousand manifest tonnes of protein, 69% of the level its route, season and drift in volume would lead one to expect. The registry's price series for protein does not show a corresponding jump. In other quarters a poor harvest at the source worlds cut arrivals without any closure at all.
Which disturbance matters more is an empirical question.
Coruscant is the galactic capital, a planet-wide city of many levels, and we take as a working premise that it is fed in large part by grain and protein carried to it from agricultural worlds along the hyperlanes. The premise is ours, and this paper neither estimates how large the imported share is nor makes any statement about the capital's total consumption. It asks only about variation in the quantities that reach the capital and in the prices they fetch.
Talk of the capital's food supply tends to invoke a disturbance of one of two types. The Republic Hyperspace Survey Office issues advisory bulletins closing a charted lane to civil traffic when the drift of a mass concentration puts transit at risk, and a closed lane stops freight as surely as it stops passengers (Republic Hyperspace Survey Office, 30–24 BBY; Ishvarne, 27 BBY). Staple returns on marginal agricultural worlds, for their part, swing from season to season with rainfall and soil condition (Hollenvar, 25 BBY). The closure and the harvest are rarely set against one another, and we found no earlier study that puts both on the same scale.
The analysis was completed in 24 BBY, the vantage of every statement below. We ask how much of the variation in quarterly arrival volume on six supply routes, and in the wholesale prices of grain and protein in the capital, can be attached to each disturbance. The answer to a question of that form depends on how shares are defined. We therefore fix the scale and the order of entry at the outset: shares are shares of within-route variance in the natural logarithm of arrivals, and of within-commodity variance in the natural logarithm of price, and each is reported with closures entered before harvests and after them.
2. Methods
Sources. Arrivals and prices come from the quarterly returns of the Coruscant provisioning registry, the receiving office that logs each consignment of grain and protein by weight and route of dispatch and records the wholesale hall price of each commodity at the close of the quarter. The returns, together with the harvest returns that source-world ministries lodge for their staple crops, were reproduced in a Senate research report (Galactic Senate Legislative Research Office, 24 BBY). Closure hours come from the advisory bulletins of the Republic Hyperspace Survey Office (Republic Hyperspace Survey Office, 30–24 BBY), which give the lane, the time of issue and the time of lifting for each closure. Both record types are treated as the paper's own compiled series, and no consignment or world is identified here.
Routes and window. We grouped the registry's consignments into six supply routes, R1 to R6, three carrying grain and three carrying protein. Each route is a chain of lane segments joining a cluster of agricultural worlds to the capital. The labels are our own and carry no administrative meaning. The window runs for 25 standard quarters, a standard quarter being one fourth of the standard year, from the first quarter of 30 BBY to the first quarter of 24 BBY. That gives 150 possible route-quarters. Three are missing because registry pages are lost (R2 in the fourth quarter of 29 BBY, R5 in the second and third quarters of 27 BBY), leaving 147, of which 74 are grain and 73 protein. The six routes have unequal sizes by the registry's own schedule, and the panel is balanced in time only because every route is reported every quarter.
Variables. Arrivals are the route-quarter's weight in thousand manifest tonnes, the weight unit of the registry's receiving manifests. The closure share is the fraction of the quarter's hours during which any advisory closed a charted lane of the route to civil traffic, with overlapping bulletins counted once. The harvest anomaly is the natural-log difference between the quarter's harvest return per hectare from the route's source cluster and that cluster's mean for the same quarter of the year over the window. For the grain routes the return is the grain crop lodged per hectare sown; for the protein routes it is the slaughter or landed catch lodged by the source ministries for dispatch, per hectare of registered pasture or fishery area, and the harvest share for protein is on that basis. Returns count what was lodged for dispatch in the quarter, which is why the anomaly enters in the same quarter as arrivals, with no shipping lag. Wholesale prices are the registry's index for each commodity in its own units, with 100 as a nominal base, a convention for such indices (Hesperin, 29 BBY). For the price analysis the closure share and harvest anomaly of a commodity's three routes are averaged with weights fixed at each route's planning volume in the registry's route schedule (40, 28 and 18 thousand tonnes for the grain routes, 14, 9 and 6 for the protein routes), renormalised over the routes present in a quarter. Closure hours are measured to the hour, whereas harvest returns are lodged by the source worlds and are good to about two figures.
Models. Arrivals were analysed by least squares on the natural logarithm of tonnage, with an intercept for each route, three quarter-of-year contrasts within each commodity, a linear trend in quarters, the closure share and the harvest anomaly. That is 15 parameters and 132 degrees of freedom. Prices were analysed in the same way with a commodity intercept, three quarter-of-year contrasts within each commodity, a trend, and the weighted closure share and weighted harvest anomaly, giving 11 parameters and 39 degrees of freedom across 50 commodity-quarters. Intervals use the t distribution with the model's residual degrees of freedom, which matters for the price model. Residuals were treated as independent, and we report their lag-1 autocorrelation within route or commodity as a check.
Shares of variance. The scale is the natural logarithm of the outcome, after removing route means for arrivals and commodity means for prices, so that differences in route size do not count as variance. We entered season and trend first. Closure and harvest were then entered one after the other in each order, and each variable's share is the fall in residual variance it produces as a fraction of the variance on that scale. Because the two variables are correlated, their shares depend on the order (Harrowane, 28 BBY), and we report both orders and their mean. Intervals for the order-mean shares and for the difference between the closure and harvest shares are percentile intervals from 2,000 resamples of whole quarters, all routes of a drawn quarter kept together (Jarrowick, 29 BBY).
3. Results
The median route-quarter delivered 15.9 thousand manifest tonnes (range 4.1–59.4). The closure share had a median of 5.2% of the quarter's hours (interquartile range 1.6–9.0%, mean 6.9%, standard deviation 0.070), and 20 route-quarters (13.6%) lay above 15%, the highest being the 31.0% described above. The harvest anomaly had a standard deviation of 0.141 natural-log units and correlated 0.11 with the closure share. The grain index ran from 115.3 in the first quarter to 141.0 in the last and the protein index from 105.3 to 114.6, with an overall range of 90.0 to 142.5. The weighted closure share had a standard deviation of 0.043 and the weighted harvest anomaly one of 0.091, correlated 0.27 within commodity.
Longer closures lowered arrivals on the route (Table 1), with a coefficient of −1.640 per unit of closure share (95% CI −1.900 to −1.381; t = −12.5, p < .001). A further ten points of the quarter closed went with a multiplier of 0.85 on arrivals (0.83–0.87), or 15.1% fewer tonnes (12.9–17.3%). Better harvests raised them: a harvest anomaly 0.10 higher went with a multiplier of 1.049 (1.036–1.062), or 4.9% more (3.6–6.2%). Arrivals drifted upward by 0.6% a quarter (0.4–0.8%). The model with route intercepts accounted for 97.8% of the variance of log arrivals, nearly all of it differences in route size, which is why the shares below are taken within route. The residual standard deviation was 0.104 natural-log units and the lag-1 residual autocorrelation 0.16.
Table 2 sets out the same relation without a model, in four bands of closure share. Route-quarters closed for less than 3% of their hours delivered 1.10 times the level expected for their route, season and trend (1.06–1.13), and those closed for 20% or more delivered 0.77 times it (0.72–0.82). The bands are not adjusted for harvest and are descriptive only.
In the decomposition of Table 3, season and trend accounted for 14.7% of within-route variance in arrivals. Closure accounted for 36.5% when entered first and 40.6% when entered after harvest, and harvest for 10.4% and 14.5% in the same two situations. Their order-mean shares were 38.5% (bootstrap 95% interval 22.6–51.2) for closure and 12.5% (5.6–21.0) for harvest. The difference was 26.1 percentage points (3.5–43.4), and closure exceeded harvest in 98.9% of resamples. The ranking held in both orders, and 34.3% of the variance remained unexplained.
Prices told a different story. The weighted harvest anomaly carried a coefficient of −0.767 (−0.955 to −0.578; t = −8.2, p < .001), so a harvest anomaly 0.10 lower went with prices 8.0% higher (5.9–10.0%). The weighted closure share carried 0.723 (0.265 to 1.181; t = 3.2, p = .003), so ten more points of closure went with prices 7.5% higher (2.7–12.5%). Season and trend accounted for 42.9% of within-commodity price variance. Harvest accounted for 30.9% when entered first and 36.0% after closure, and closure for 0.3% first and 5.4% after harvest. Order-mean shares were 33.4% (13.9–44.9) for harvest and 2.9% (0.5–11.4) for closure, a difference of −30.6 points for closure minus harvest (−42.2 to −8.9). Harvest exceeded closure in 99.9% of resamples, and 20.8% of the variance was left over. The ratio of the harvest share to the closure share was 5.7 with harvest entered first and above 100 with closure entered first (118 on the unrounded shares). The price residuals had lag-1 autocorrelation of 0.17 and a standard deviation of 0.053 natural-log units.
Two further comparisons bear on the contrast between the models. Averaging the three routes of a commodity cut the closure standard deviation from 0.070 to 0.043, leaving about 37% of its variance, and the harvest standard deviation from 0.141 to 0.091, leaving about 42%. The weighting therefore shrank the two predictors by almost the same proportion. The coefficients differ more. Per natural-log unit of arrivals lost, a harvest shortfall went with a price rise of 0.767/0.478 = 1.60 natural-log units, and a closure with one of 0.723/1.640 = 0.44, a ratio of 3.6. The price coefficients are per unit of a three-route average, so this comparison is rough.
Measurement error in the harvest anomaly would lower its share. Under classical error the share falls roughly in proportion to the reliability of the measure, so equalising the two arrival shares would take a reliability of 0.32 (12.5/38.5). Returns good to about two figures carry a rounding error of at most about 5%, or 0.05 natural-log units. If rounding were the whole of the reporting error, the reliability against the anomaly's standard deviation of 0.141 would be at least 0.87. That assumption is untested: lodging errors, and the area denominators of the protein routes, are not covered by it. Equalising the two shares would take a reporting error with a standard deviation of about 0.12 natural-log units, more than twice that bound.
The swing in the closure share between orders for prices, from 0.3% to 5.4%, shows how sensitive the apportionment is to order when predictors are correlated.
4. Discussion
Within a route, the records indicate that advisory closures account for roughly three times as much of the variation in arrivals as harvests do, and that the order of the two reverses for the capital's wholesale prices, where harvests dominate in either order (by a factor of about six with harvest entered first, and by far more with closure entered first). We consider both rankings likely to hold, since they survive both orders of entry and the resampling of quarters. The size of the closure effect on arrivals is large in plain terms: a quarter in which a fifth or more of the hours are closed delivers about a quarter less than the route's expected level, on the band means of Table 2.
The ranking for arrivals needs an alternative explanation set beside it.
Closure hours are counted to the hour from bulletins, whereas harvest returns are lodged by the source worlds and are coarser, and measurement error in harvest would shrink its share. We think this unlikely to account for the whole gap. The same harvest measure carries a third of the variance in the price series, and its coefficient on arrivals is estimated with an interval of 0.351 to 0.604. The better reason is one of size: equalising the two arrival shares would take a reliability near one third, which means a reporting error of about 0.12 natural-log units, more than twice the 0.05 that two-figure rounding allows (Results). Unless the source worlds' lodging errors are far larger than their rounding, attenuation of the needed size is improbable, so the ranking for arrivals is more probably real than an artefact of the instruments, though a modest understatement of the harvest share cannot be excluded.
What stands out is the contrast. Closures remove about 15% of a route's tonnage per ten points of the quarter lost, yet they add little to the price variation the capital experiences once harvests are known. Averaging three routes dilutes both predictors almost equally, so the weighting accounts for little of the contrast. The coefficients carry it: per unit of arrivals lost, a harvest shortfall went with a price rise about 1.6 times as large in log terms, a closure with one about 0.44 times as large. The difference may reflect diversion of consignments to routes that stayed open, or stock drawn in the capital, the usual buffer between supply shocks and wholesale prices (Ioranne, 27 BBY), and neither can be examined here, because the returns carry no stock series and no consignment-level record of rerouting.
A positive closure coefficient for prices, clear of zero at 7.5% per ten points, shows that closures are not costless to the capital, only small beside the harvest. It is possible that the price effect would be larger in a quarter with simultaneous closures on several routes, a configuration that is rare in a window with only six routes.
5. Limitations
The records are observational and archival. Closures follow the drift of mass concentrations, which no carrier controls, and that helps, but harvests at the source worlds may correlate with unrecorded shocks to the same routes, and our models contain no term for them. The window ends in the first quarter of 24 BBY and says nothing of later quarters.
Six routes is a small number. The route intercepts absorb everything fixed about a route, but the slopes are common to all routes, and a route-specific slope cannot be estimated with this many. The price models rest on 50 commodity-quarters, two commodities pooled, and their bootstrap intervals are correspondingly wide: the closure share of price variance could lie anywhere from 0.5% to 11.4%. Residual autocorrelation is mild (0.16 for arrivals, 0.17 for prices), but the intervals treat residuals as independent and are therefore probably a little short.
None of these shares measures what the capital's households ate.
A closure share records the hours a lane was closed by advisory, not the traffic that was actually turned back, and a carrier that waited out a closure at the lane's mouth is counted the same as one that never set out. The registry sees only consignments logged through its receiving office, so goods that reached the capital by other channels are absent. We also have no figure for the capital's stocks, so the arrival and price results cannot be joined into a single account of supply.
The shares themselves are conventions. They are on the log scale, within route or commodity, and they depend on the order of entry; with correlated predictors there is no unique division. The order-mean is one defensible summary, and the two orders are shown in Table 3 so that a reader can choose another. Whether any division of this kind would survive a design that observed stocks and rerouting directly is an open methodological question for records studies of provisioning, and the registry's returns, in their present form, cannot settle it.
References
- Galactic Senate Legislative Research Office (24 BBY). Quarterly returns of the Coruscant provisioning registry and source-world harvest returns, grain and protein, first quarter of 30 BBY to first quarter of 24 BBY. Galactic Senate Legislative Research Office Reports, Report LRO-24/12.
- Republic Hyperspace Survey Office (30–24 BBY). Hyperlane advisory bulletins on mass-concentration drift, quarterly compilations, supply routes to Coruscant. Republic Hyperspace Survey Charts, Advisory bulletin compilations, quarters 1 to 25.
- Ishvarne, P. (27 BBY). Estimating the drift of catalogued mass concentrations from successive lane advisories. Duro Astrogation Institute Transactions, 63(2), 71–94.
- Hollenvar, R. (25 BBY). Rainfall failure and the interannual variance of staple returns on marginal agricultural worlds. Journal of Arid-World Ecology, 19(1), 3–30.
- Ioranne, T. (27 BBY). Stockholding and the pass-through of supply shocks to wholesale staple prices. Muunilinst Economic Analysis Bulletin, 16(1), 33–58.
- Hesperin, K. (29 BBY). Wholesale hall price indices for staple goods, construction and the choice of base period. Journal of Galactic Public Finance, 11(2), 61–84.
- Harrowane, D. (28 BBY). Order dependence in sequential variance shares for correlated predictors. Proceedings of Applied Speculative Statistics, 9(1), 12–35.
- Jarrowick, S. (29 BBY). Resampling whole periods in balanced panels of archival records. Proceedings of Applied Speculative Statistics, 8(4), 171–192.
Open in Uncited Press →