silk trade between ancient Rome and China

About two millennia ago, silk clothes were fashionable in Rome.  Moreover, a leading Roman physician had in his personal library in Rome about 192 GC “books written on white silk, with black covers, for which he had paid a high price.”  It’s likely that these books were silk scrolls.  In any case, the Romans clearly had a considerable amount of silk goods.

Evidence of considerable silk trade between ancient Rome and China is less clear.  Roman texts indicate that silk came from “Seres,” which means “the land of silk.”  Seres has subsequently been interpreted as China.  A merchant’s trading manual from first-century Egypt provides a more direct connection between silk and China:

Beyond this region {the mouth of the Ganges River}, by now at the northernmost point, where the sea ends somewhere on the outer fringe, there is a very great inland city called Thina {China} from which silk floss, yarn, and cloth are shipped by land via Bactria to Barygaza and via the Ganges River back to Limyrike.  It is not easy to get to this Thina; for rarely do people come from it, and only a few.  The area lies right under Ursa Minor and, it is said, is contiguous with the part of the Pontus and the Caspian Sea where these parts turn off, near where Lake Macotis, which lies parallel, along with {the Caspian} empties into the ocean. [1]

This text’s geography is muddled.  It suggests, however, that some, but not a lot, of silk trade may have existed between China and ancient Rome.  Oasis economies in central Asia show little evidence of long-distance silk trade.[2]

Perhaps ancient Romans got a large volume of silk goods from some place other than China.   Extraction of silk from wild silkworms for use in textiles has occurred in South Asia outside of China since the late third millennium BGC.  Moreover, that silk in some instances appears to have been reeled, a relatively advanced silk-processing technique.[3]  The domestication of silk worms was a secret apparently exclusive to the Chinese until after the collapse of the western Roman Empire.   Perhaps silk goods in ancient Rome were somewhat lower quality silk goods made from wild silk worms processed outside of China.

Perhaps silk goods in ancient Rome came from China via a route different from what has been called for just over a century the “silk road.”  The cultural and economic sophistication of loose networks of nomadic peoples has tended to be underestimated.  Silk trade between Rome and China may have been organized through nomadic Central Asian peoples, rather than via a “silk road” connecting oasis settlements.  Another alternative to the “silk road” is maritime trade.  Maritime routes from western Eurasia to China are thought to have been established from the eleventh century.  Perhaps maritime trade between the eastern and western ends of Eurasia was established much earlier than previously thought.

No authentic Roman-era coins have been found in China.[4]  That suggests that if trade between ancient Rome and China occurred, it occurred within a network of transactions that separated the circulation of Roman coins from the transfer of silk from China to Rome.  Whether ancient Rome got silk from China or elsewhere, the ancient economy of Eurasia was more complex than the exchange of silk for gold between China and Rome.

Tarim mummy yingpan has silk clothing


*  *  *  *  *

Related posts:


[1] From the Periplus Maris Erythraei, trans. Casson (1989) p. 91. An alternate translation is freely available online.

[2] Hansen (2012).

[3] Good, Kenoyer & Meadow (2009).

[4] Hansen (2012) p. 20.


Casson, Lionel. 1989. The Periplus Maris Erythraei: text with introduction, translation, and commentary. Princeton, N.J.: Princeton University Press.

Good, Irene, J. Mark Kenoyer, Richard H. Meadow.  2009. “New Evidence for Early Silk in the Indus Civilization.” Archaeometry. 51 (3): 457-466.

Hansen, Valerie. 2012. The Silk Road: a new history. Oxford: Oxford University Press.


Google Fiber gets good government regulation

Google Fiber has set up for itself a relatively good framework of government regulation in Kansas City.  Some regulatory advantages of Google’s choice of Kansas City:

  1. Google Fiber chose Kansas City from among more than 1000 cities that actively sought Google Fiber.  Kansas City’s actively expressed interest in getting Google Fiber indicates that the city governments will be relatively amenable to Google Fiber’s needs.
  2. Google isn’t providing telephone service.  Telephone service is predominantly regulated at the state and federal level.  Selecting among interested cities does less to get a favorable state regulatory framework.
  3. Kansas City actually consists of two adjacent city jurisdictions — Kansas City, Missouri; and Kansas City, Kansas.  Google Fiber thus has two chances for good regulation.  Moreover, it can easily compare the effects of regulation across jurisdictions.  In addition, it can shift its network build across jurisdictions if regulatory differences (local and state) become highly significant.

Government regulation is one among many challenges in building and operating communications networks.  Just as competition among businesses to serve customers typically benefits customers, competition among governments to encourage broadband typically benefits broadband providers.  If governments compete wisely, that competition will also benefit the public.

*  *  *  *  *

Related posts:


Google Fiber confronts geographic heterogeneity

Google Fiber‘s network build in Kansas City is publicly organized around fiberhoods.  A fiberhood is a small, geographic area that is a candidate for having Google fiber installed.[1]  Google is prioritizing its network build across fiberhoods based on the extent to which fiberhoods exceed pre-registration goals.  Differences in pre-registration goals across fiberhoods suggest cost and expected per-household profitability heterogeneity across fiberhoods.  Differences in pre-registrations received indicate take-up heterogeneity across fiberhoods.

The initial distribution of pre-registration goals across fiberhoods was simple.  A majority of fiberhoods (57%) had a preregistration goal of 10%.  Only two other pre-registration goal levels existed: 25% (for 13% of fiberhoods) and 5% (for 29% of fiberhoods).  The distribution of network build costs across fiberhoods is probably much more continuous than this simple pre-registration goal distribution.  Hence a reasonable conjecture is that the pre-registration goals, which differ by a factor of five, are not closely related to network build costs.  Heterogeneity in expected per-household profitability, or some other factor, apparently is determining pre-registration goal levels.[2]

Pre-registration take-up heterogeneity appears to be significantly affecting Google Fiber’s build.  As of August 29, 2012, about 15,900 households had preregistered, and about 99 fiberhoods (out of 202) had met or exceeding their preregistration goals.  Each fiberhood exactly meeting its pre-registration goal implies about 16,200 pre-registrations.  Hence Google Fiber has nearly met in aggregate its pre-registration goal, but the distribution of those preregistrations implies a network build to less than half of the fiberhoods.  Google Fiber recently announced that it has adjusted downward its pre-registration goals. That change favors a broader network build.

Successfully investing in mass-market network infrastructure is a difficult business.  Four cable companies have unsuccessfully attempted to build networks in Kansas City.   According to public FCC data, the number of providers of high-speed Internet access varies considerably across census tracts in Kansas City.  Most high-speed internet service provides in Kansas City probably serve only a few, highly specialized customers.  Google Fiber is an interesting mass-market experiment that is likely to evolve as its reveals more information about network costs and customer demand.

*  *  *  *  *

Data: public data on Goggle Fiber fiberhoods in Kansas City, as well as public data on other networks (Excel version)


[1]  Currently defined for the first stage of the Google Fiber build in Kansas City are 202 fiberhoods in Kansas City, Kansas, and in Kansas City, Missouri.  The median fiberhood size is 771 households, with upper and lower quartiles of 1096 and 461 households.  For details, see the fiberhoods sheet.

[2] As of Aug. 29, the median pre-registration shares across the sets of fiberhoods with goal shares of 5%, 10% and 25% were 8%, 6%, and 13%, respectively.  If goal shares are inversely correlated with household income, that pattern of take-up suggests free Internet service is dominating take-up.


wireless network advantage


Wireless networks never require retrenchment.


interactive maps of broadband speeds

A recent UK online newspaper article includes an interactive map of UK broadband speeds.  Text above the map states, “Use our interactive map to find the results for your area.”  That statement presents the map as a data access tool.  But the map includes color-coded markers that provide a quick visualization of the geographic distribution of broadband speeds.  Unfortunately, hard-coded into this visualization are particular color-change levels:

  • Large red icon: Under 4Mbps
  • Small yellow icon: Between 4Mbps and 7Mbps
  • Large green icon: Over 7Mbps

The article provides no explanation for the use or importance of 4Mbps and 7Mbps thresholds.  Given that the interactive map doesn’t allow the user to change these thresholds, at least some explanation should have been provided to justify those choices.

With Needle, users can easily change mapping thresholds.  Here’s a broadband speed map for the US, and here’s how to change the mapping threshold.


relatively slow reduction in bandwidth prices

Quality-adjusted average U.S. residential broadband service prices have fallen no more than an estimated 10% from 2004 to 2009.[1]   The consumer price index for personal computers and peripheral equipment fell 50% across that period.[2]   The price-performance frontier for communications technology is advancing as fast or faster than that for personal computers and peripherals.  The difference in realized price trends reflects much different structures of investments, transactions, and business competition.

*  *  *  *  *


[1] See Greenstein, Shane M. and McDevitt, Ryan C., Evidence of a Modest Price Decline in US Broadband Services. Center for the Study of Industrial Organization, Northwestern University, Working Paper #0102 (January 2010).  The bandwidth figures in this paper are mistakenly labeled “bps” (bits per second).  They actually are in “kbps” (kilobits per second).  I found that average wholesale local bandwidth prices fell about 20% from 1990 to 1995, and remained roughly constant from 1995 to 2000.  See Galbi (2000), “U.S. Bandwidth Price Trends in the 1990s,” Table P4.  For related discussion, see Galbi (2000), “Growth in the ‘New Economy’: U.S. Bandwidth Use and Pricing Across the 1990s.”  All these reported figures are nominal, i.e. not adjusted for general price inflation.

[2] U.S. Bureau of Labor Statistics, Consumer Price Index Detailed Report, Aug. 2010, Table 21.


mapping average Internet download speeds in U.S.

Akamai’s publicly filed dataset of observed average Internet download speeds for U.S. network connections is now on Needle.  Needle makes it easy to map the reported cities and states according to an average speed threshold.  For example, 79 out of the 500 reported cities have average Internet download speed less 2048 kbit/s.  Here’s a map of states that contain those cities.  The map shows that relatively low average Internet download speeds by city are widely dispersed geographically.

To generate a map for an average speed threshold of your own choosing, click on “city” under “every” on the left of the Needle page.  Then click on “>>” on the table heading, just to the right of “every     city.”   Then click on “keep only those where…”, and then click on “ave speed.”  Ave speed jumps to the top of the list and a drop-down selection box appears.   Select “is less than,” and enter a figure (kbit/s) in the box to the right.   Then click on next.  The screen will change slightly to reflect your choice, and a “done” button will appear.  Click on the “done” button.   You’ll then see a list of the relevant cites.  On the upper right, change the drop-down selector from “table” to “map.”  You’ll then see a map with markers for states that contain cities that satisfy your selection criterion.  If you change the “group by” drop-down selector from “state” to “city,” the map will change to include markers for each city that satisfies your selection criterion.  You can similarly map state average speeds by clicking on “state” rather than “city” in the first step.  To change your selection criterion, you must start by clicking on “city” or “state” on the left and redo the steps above.

Related post: dispersion of Internet download speeds


dispersion of Internet download speeds

Better Internet connectivity tends to be associated with more urban areas, areas with a greater concentration of high-tech industries and employees, and areas with wealthier, more educated populations.   These factors, however, do not provide simple explanations for the actual geographic pattern of Internet download speeds from Akamai’s server network.  According to Akamai’s measurements (which include residential and business customers), the U.S. state with the highest average Internet download speed in the second quarter of 2009 was New Hampshire.  New Hampshire is noted for extensive forests, beautiful mountains, and ice fishing.  Illinois, in contrast, includes Chicago, the third-largest U.S. city and long a major hub of trading and banking.  In average Internet download speed, Illinois ranks 45 out of all 51 U.S. states and the District of Columbia.  Illinois’ average Internet download speed is only 46% that of New Hampshire.  While New York state is near the top of the average speed ranking and Alaska is at the bottom, unexpected relative positions, such as those of New Hampshire and Illinois, are prevalent in the ranking.

Unexpected dispersion in Internet download speeds appears in other Akamai data.  Looking at the distribution of download speeds across IP addresses within states, Washington state, which includes the headquarters of Microsoft and other high-tech companies, has among the lowest shares of IP addresses downloading at faster than 768 kbits/s.    That share for Washington is 77%.   Nevada and Maine, in comparison, have 98% and 96%, respectively, of IP addresses downloading at faster than 768 kbits/s.[1]   Looking at download speeds by cities, the city with the highest average download speed is Sandy City, Utah, and the next highest, Norman, Oklahoma.[2]   Most persons have never heard of either.

Dispersion in Internet download speeds suggests that idiosyncratic organizational factors greatly affect Internet connectivity.[3]   Technology for providing relatively high-speed Internet access is well understood and widely available.   But Internet connectivity impinges on a vast array of organizational activities and interests.  That’s a real Internet congestion problem.

*  *  *  *  *

Data: Internet download speeds across U.S. states and cities, as measured by Akamai (Excel version)


[1]  For the U.S. as a whole, the FCC’s OBI Technical Paper No. 4, Broadband Performance, shows that 88% of U.S. Internet users have actual download speeds greater than 1 Mbps.  See Exhibit 18, which is based on comScore data for the first half of 2009.   Few comScore data are publicly available and little is know about the specifics of comScore’s measurements.  See Steve Bauer, David Clark, and William Lehr, “Understanding Broadband Speed Measurements,” pp. 16-7.   In the UK in May, 2010, about 92% of residential broadband connections had actual average download speeds greater than 4 Mbps.  See UK Broadband Speeds 2010.  Estimate based on Figures 4.2 and 4.5.

[2]  The set of cities considered are the top-ten cities by IP address density in each state.  See Akamai, “Observed Average Internet Speeds for U.S. Network Connections,” p. 2.

[3] This dispersion does not particularly characterize the U.S.  Considering mobile broadband world-wide in 1Q 2010, Akamai observed:

we see that there is an extremely wide range in average connection speeds – oddly enough, the highest (7175 Kbps) and the lowest (105 Kbps) were both seen on providers in Slovakia.

See Akamai, State of the Internet, 1st Quarter, 2010 Report, p. 25.


online database of DS1 and DS3 special access rates

The DS1 and DS3 rates that the Ad Hoc Telecommunications Users filed publicly at the FCC are now accessible as an online, highly capable Needle domain (database).  Needle is a data system that makes it easy to look at the data in different ways and to sort and filter it, all from within a web browser.

The original filings (here, here, and here) provide the data as pdf pages displaying tables with highly complex row and column structures.   A human can read and page through the data as if it were text.  That data format serves neither the reading capabilities of humans nor the data-processing capabilities of computers.

To make the Ad Hoc DS1/DS3 rate data more accessible,  I extracted it from the pdf files and re-organized it into one, regular, comma-separated-value (CSV) file with 3698 data rows.  I also put together some relevant data documentation.   Analyzing the CSV file with a spreadsheet is possible but cumbersome.   Since the CSV file has a simple tabular data form, it’s easy to analyze with a database program, if you have one.   You would download the data, import it into the database program, and then set up and run a query that generated the data view that you seek.

Needle makes many different views of the data easily accessible to a web browser.  Within Needle, a dataset is a graph of data nodes, where each data node is a single piece of data of a particular type.   The Needle Ad Hoc DS1/DS3 domain shows (on the left under “Every:”) a linked list of every node type in that dataset.   If you click on any of these node types, a table will appear that has as its leftmost column a list of all the data nodes of the clicked type.   So, for example, if you click on “bandwidth,” you will see the nodes DS1 and DS3 in the left column of the table.  The table also shows the number of attribute sets and the average circuit10 rate (a composite rate) across the DS1 and DS3 nodes, respectively.  You can look at the circuit10 rates by clicking on the circuit10 link (node type) on the left.   The resulting table shows all the circuit10 rates, in descending order, in the left column.  Other columns of the table show other attributes associated with each circuit10 rate.

For any table that you see, you can filter, sort, and group the data.  For example, to limit the table of circuit10 rates to DS1 rates, left click on the “bandwidth” column heading, select “filter by this column” in the pop-up menu, type DS1 into the box next to “show”, and then click on “do” just to the right of that box.   The table will then contract to show just the DS1 circuit10 rates.  A similar procedure produces filters for company, year, state, reg type, term, and zone.  If you want to see the elements of each of these data types, click on that type on the left.   Options on the pop-up menu also provide for sorting and grouping.  Under “Index” on the top left, the “rates” and “rates subset” links show examples of tables made from grouping, filtering, and  sorting the cn (attribute set) nodes.  The “compare 2009 to 2006″ and “compare 2009 to 2005″ links under the index heading show tables that include circuit10 price ratios across the relevant years.  You can sort and filter these tables like any other table.

Any subset of data can be extracted easily from Needle.  At the bottom of each table are links “See this data as: Plain List · CSV · JSV · JSONa”.  Just click on CSV to download a CSV file of the data.  If the table has groups, you need to flatten the table (switch grouping to a regular data column) before exporting.  Needle also offers API functionality that allows Needle to serve as a data repository for high-powered statistical analysis packages such as R or S.

Needle can do much more than what it is doing for the Ad Hoc DS1/DS3 dataset.  Needle’s strengths include data acquisition, merging, and cleansing.  In addition, Needle’s graph-based data organization can easily handle complex data structures that create nightmares in traditional relational databases, which require tabular data forms.  Needle, for example, can easily handle variable-length lists of items.  None of these strengths are applied to present the Ad Hoc DS1/DS3 dataset.   Needle here merely makes the Ad Hoc DS1/DS3 data much more easily accessible, especially compared to data published as pages of tables in a pdf document.


DS1 & DS3 rate dispersion across U.S. states

Based on data filed by the Ad Hoc Telecommunications Users Committee, tariff rates in 2009 for DS1 and DS3 special-access circuit elements across U.S. states have a spread equal to about plus and minus a third of the average.  Rates differ across bandwidth (DS1 or DS3), regulatory type (price cap or pricing flexibility), purchasing term commitment (in months from 1 to 60), and geographic zone (typically three zones).  Differences across states within these rate structures reflect other factors that affect tariff rates.

Rate differences across states are not highly correlated with state characteristics.  Qwest, for example, has the same DS1 and DS3 rates across its 14-state service territory.  AT&T and Verizon, in contrast, have tariffs that differ across state groups in ways that relate to the service territories of historic telephone operating companies.

Consider the highest and lowest DS1, price-cap, month-to-month, zone 1 rates as measured by the composite 10-mile circuit rate.  The highest such rate is $1023 in Indiana and Wisconsin (AT&T).  The lowest such rate is $395 across the whole 14-state Qwest service territory, which includes Minnesota and Iowa.   Differences in regulation, competition, service cost, or unmeasured differences in tariff structures could explain this dispersion.  What specifically explains the actual difference isn’t obvious.

Differences between price-cap and pricing-flexibility rates also show considerable ambiguity.  Telephone companies are granted petitions for pricing flexibility based on criteria that the FCC established to measure the development of competition.[*]  The rate data indicate that pricing flexibility rates are consistently higher than price-cap rates.   Higher prices typically aren’t associated with greater competition.  However, for most service attribute types, pricing flexibility rates have less dispersion across states than do price-cap rates.  That is consistent with more competition in circumstances in which unpriced differences across states matter little.

*  *  *  *  *

Data: DS1 and DS3 rate statistics based on Ad Hoc Rate Dataset (Excel version); Ad Hoc DS1 and DS3 Rate Dataset


[*] An FCC order, adopted on Aug. 5, 1999, set out a procedure (“pricing flexibility” petitions) for removing rate elements from existing price-cap regulation.  BellSouth provides an example of the regulatory procedure.  On Dec. 15, 2000, the FCC’s Common Carrier Bureau granted a BellSouth petition for pricing flexibility.  The order granting that petition apparently isn’t online, but an affirming review of that order, which includes a list of the metropolitican statistical areas (MSAs) to which it applies, is online.   Here’s a better formated version of the MSA list.  On Nov. 22, 2002, the Bureau adopted an order granting another BellSouth petition for pricing flexibility.  On May 16,2008, the Bureau granted a third BellSouth petition for pricing flexibility.


Next Page »