I wish to thank Professors Rudolph Bell, Alexander Kazhdan, Walter King, John Trierweiler, and Dr. Paul Hollingsworth for reading and commenting upon earlier versions of this paper. A preliminary treatment of this topic was read at the 17th International Byzantine Congress in Washington.

1. Since the sources themselves provide very little direct information on the daily workings of the imperial chancery, previous scholarship has, by necessity, focused on diplomatics and the general administrative structure of the imperial bureaucracy. On imperial documents, see Dôlger, F. and Karayannopulos, J., Byzantinische Urkundenlehre, I: Die Kaiserurkunden (Munich 1968)Google Scholar, and for further bibliography, Karayannopulos, J. and Weiss, G., Quellenkunde zur Geschichte von Byzanz (324–1453) (Wiesbaden 1982) 91–93, 100–02 Google Scholar. On the chancery, Dölger, F., ‘Die byzantinische und die mittelalterliche serbische Herrscherkanzlei’, Actes du Xlle Congrès international d’études byzantines (Belgrade 1963) I, 83–103 Google Scholar; Oikonomidès, N., ‘La chancellerie impériale deByzance du 13e au 15e siècle’, REB 43 (1985) 167–195 CrossRefGoogle Scholar; and Loenertz, R.-J., ‘Le chancelier impérial à Byzance au XlVe et au XHIe siècle’, OCP 26 (1960) 275–300 Google Scholar (primarily on the ).

2. Dölger provides a more extensive explanation of his methodology in the introduction to each part of the Regesten.

3. See Dólger and Karayannopulos, Byz. Urkundenlehre, 84 and 98.

4. Vranouse, E., , 1 (Athens 1980)Google Scholar no. 19. For katastrosis in the Lavra dossier, see Lemerle, P. et al., Actes de Lavra I (Paris, 1970), index.Google Scholar

5. On the general subject of the datation of documents, see Oikonomidès, Chancellerie impériale, 180–87, who points out that, at least in the late period, the datation passage of documents provides only an approximate indication of when the emperor actually signed the act, and Dölger and Karayannopulos, Byz. Urkundenlehre, 49–54.

6. Before the decision is made to use monthly data, one must make reasonably sure that there was no tendency to issue imperial documents at a particular time of the month. For this, an arrangement by day of month for these 158 documents was useful. While some particular days thus seem to have been popular, and for good reason, such as the 1st (twelve documents, or 7.6%) and 25th of the month (eleven, or 7.0%), and others less popular, such as the 16th and the 27th (both with only one document, or 0.6% each), no consecutive period of fifteen days in the daily array embraces significantly more than 50% of the documents. This is an indication that the results of a monthly array would not be distorted by activity cycles within months.

7. The problem lies in determining a methodologically-sound dividing line in such cases. A document issued ‘bald nach April 6.’ could reasonably be assigned to April, but what of a document issued ‘bald nach April 29.’ or ‘bald nach April 25.’, or even ‘vor April 25.’? Any criterion for sorting such documents by month would be both arbitrary and subjective.

8. In other words, while the 861 documents in the data set are far more than is needed to construct a valid seasonal index, a graph plotting the number of documents per month for the 888 years of the Regesten has no seasonal patterns discernible to the eye, unlike, say, a graph of monthly ice cream sales which would regularly show lower values in colder months. For similar reasons, the data is unsuitable for the use of simple autocorrelation techniques.

9. The analysis used the version of the FFT included within the SYSTAT statistical package (on the fast Fourier transform, see, e.g., Bloomfield, P., Fourier Analysis of Time Series: An Introduction [New York 1976] 61ff.)Google Scholar. In order to simplify encoding the data and to accommodate the memory limits of the program, the actual time series, from September 565 to May 1453 and comprising (887 years x 12 months) + 9 months = 10653 months or periods, was compressed by dividing the data into ten-year intervals beginning with September 565-August 574, and then creating a single twelve-month array for each ten-year interval by summing the individual monthly number of documents across each of the ten years (e.g., September 565 + Sept. 566 +… + Sept. 574). The effect of this is slightly to intensify any seasonal periodicity present in the data; if no seasonality is present in the data, the procedure has no effect on the analysis.

This transformation of the data produced a series consisting of 1056 periods (88 ten-year periods x 12 months) incorporating data from September 565 through August 1445 (since only whole ten-year periods were used, the last eight years of data [twenty documents] were excluded). This new series was then detrended (using the least squares method), the mean of the series was subtracted from each period, and the resulting series was lengthened with zeros to 2048 periods (a power of two as required by the FFT). The FFT of this transformed data showed dominant peaks at ordinates 8, 3, 15, 6, 22, and 170 (in order of decreasing magnitude). In this case, ordinate 170 corresponds to a twelve-month periodicity (i.e., 2048 ÷ 170 ≈ 12 months).

10. There are numerous methods for calculating a seasonal index. All involve manipulating the data so that years of greater or lesser activity are not over- or under-represented in the final index. This is why merely grouping the documents by month (i.e., totalling the number of documents issued in all Januarys, then in all Februarys, etc.) does not provide a valid seasonal index, because it over-represents years for which we have many documents overall and under-represents years with very few documents. Rather, the method employed in this analysis was the ratio-to-moving-average method based on percentages of a centered twelve-month moving average for each month. In this procedure, a twelve-month moving average was first computed on a time series consisting of the number of documents issued per month per year for the time periods under consideration. (The purpose of the moving average is to isolate the seasonal component, i.e., the difference between the original time series and thç smoothed time series produced by taking a 12-month moving average is the seasonal component.) Since the moving average often equalled zero because numerous twelve-month periods contain zero documents, a constant of.01 was added to each monthly period in order to avoid dividing by zero. This procedure, which has no practical effect on the relation between the resulting seasonal index values, was necessary because the next step requires that the number of documents for each monthly period be divided by the moving average figure for each monthly period, yielding percentages of the centered twelve-month moving average. Next, after eliminating extreme values, the percentages were meaned for each month, and a seasonal index was calculated by dividing the means of these percentages by the mean of the twelve meaned percentages. The formula is

for m = 1,…,12,

where sn_m is the seasonal index, and sn_m are the averages of the arrayed monthly yt/M.A.. values (adjusted so that values more than ±1 SD from the means of the adjusted monthly arrays of yt/M.A. are excluded), where yt is the number of documents for each monthly period t = l,…, n, and M.A. is a centered twelve-month moving average of yt. On this well-known technique, commonly used to evaluate economic data, see, for example, Bowerman, B. and O’Connell, R., Time Series and Forecasting (North Scituate, Mass. 1979) 224–8 Google Scholar; Croxton, F.E. et al., Applied General Statistics, 3rd ed. (Englcwood Cliffs, N.J. 1967) 290–93 Google Scholar; or Croxton, F.E. et al., Practical Business Statistics, 4th ed. (Englewood Cliffs, N.J. 1969) 351–55 Google Scholar. The procedure requires that each document be assigned to a specific month within a specific year. This is why thirty-one documents (an extra 4%) for which Dölger assigns only a firm month have been excluded from the data set.

11. One may note that the period from 885 to 1015, while consisting of only twenty-one documents and exhibiting (via an FFT) only a rather weak seasonal factor, has a seasonal index that correlates with the index for 1015–1285 at a significance beyond the.02 level. This leads me to suspect that seasonality played a role in the frequency of document issuance prior to the eleventh century and that the actual seasonal index of these early centuries was probably similar to that of the period 1015–1285.

12. On the difficulties of winter travel, see Kazhdan, A. and Epstein, A., Change in Byzantine Culture in the Eleventh and Twelfth Centuries (Berkeley 1985) 48–49.Google Scholar

13. Nevertheless, from a cursory examination of Anna Komnena and Kinnamos, I have not been able to identify a single document (aside from the occasional peace treaty or piece of diplomatic correspondence) that was unquestionably issued while Alexios I, John I, or Manuel I was on campaign. However, since the chronology of most of these campaigns is far from sure, no real significance should be attached to this observation, and the question of whether emperors issued documents while on campaign, that is, whether they took their chancery with them, remains an open one. Common sense suggests that this might depend upon the particular emperor, whether the campaign had been carefully planned or was an emergency response, and the distance and duration of the campaign.

14. On salary distributions, Lemerle, P., ‘“Roga” et rente d’État aux Xe-XI siècles’, REB 25 (1967) 77–100 Google Scholar, esp. 95, and Hendy, M.F., Studies in the Byzantine Monetary Economy c.300–1450 (Cambridge, 1985) 160.Google Scholar

The month in which Easter falls does not appear to have an effect on the relative number of documents issued in March and April. During the period 1015–1285, Easter fell in March in 26.3% of the years, and for this period the data include thirty-seven documents issued in March, of which nine, or 24.3%, were issued in years with March Easters. Similarly, of the forty-six documents issued in April, fourteen, or 30.4%, were issued in years with March Easters. For this data, x² =0.47 (where even the.10 level of significance requires a x² of 2.71), and thus, evidence is insufficient to conclude that the month in which Easter occurred affected the number of documents issued in March and April. While this procedure was not performed on any of the other movable feasts, one may infer that the results would be similar, since all of them are dependent on the date of Easter.

15. Although such a hypothesis may seem likely, if not self-evident, on the basis of common sense, I have been able to find little in the scholarship that points in this direction. The standard works on the court and daily life in Byzantium, such as Ebersolt, J., ‘Études sur la vie publique et privée à la cour byzantine’, Revue de l’histoire des religions 26 (1917) 3–105 Google Scholar; Oeconomos, L., La vie religieuse dans l’Empire byzantin au temps des Comnènes et des Anges (1918: repr. New York 1972); Tamara Talbot Rice, Everyday Life in Byzantium(London 1967)Google Scholar; and Walter, G., La vie quotidienne à Byzance au siècle des Comnènes (1966)Google Scholar, say little, if anything, on the subject of summer excursions to the countryside. Rice, op. cit., does mention villas and hunting lodges outside Constantinople (p.47), as well as Antioch’s summer resort at Daphne (p. 141). To my knowledge, no one has ever suggested a particular month for excursions, ‘vacations’, hunting trips, etc. It is, I think, precisely an issue like this where quantitative analysis is useful. No doubt there are many passages throughout the histories and other sources that speak of summer excursions and their month, but they are scattered far and wide throughout the texts, and the researcher tends to ignore them, unless there is some indication that noting them might be worthwhile. In this case, quantitative analysis provides that indication.

In passing, I note two fateful hunting trips, for which we know their month of occurrence: Basil I’s fatal hunting accident took place in late August (Vita Euthymil patriarchae CP, ed. Karlin-Hayter, P. [Brussels 1970] 3)Google Scholar. John II died on April 8, also the result, according to Kinnamos, of a hunting accident, though, indeed, he was campaigning in Cilicia at the time (Ioannis Cinnami Epitome, ed. Meineke, A. [Bonn 1836],24–29).Google Scholar

16. See, for example, the legislation of Leo VI (J. and Zepos, P., Jus graecoromanum I [Athens 1931] 155)Google Scholar and of Andronikos II (ibid. 568–579; Dölger, Regesten, no. 2196). On the latter, see Grumei, V., ‘Le mois de Marie des Byzantins’, EO 31 (1932) 257–269 Google Scholar. Cf. Zepos, , Jus, I, 535–36 (Dölger, Regesten, no. 2295).Google Scholar

17. See, for example, Manuel I’s law reducing the number of holidays on which the law courts had to be entirely closed: Zepos, , Jus,I, 397–402 Google Scholar; Kinnamos, 276–77; Dölger, Regesten, no. 1466.

18. Oikonomidès, N., Les listes de préséance byzantines des IXe et Xe siècles (Paris 1972) 164–225 Google Scholar; Constantini Porphyrogeniti imperatoris De cerimoniis aulae byzan-tinae libri duo, ed. Reiske, I., I (Bonn 1829) 22–190, and 284–377 Google Scholar (= Constantin VII Porphyrogénète, Le livre des cérémonies, ed. Vogt, A. [Paris 1935-1940] I, 17–178, and II, 94–179)Google Scholar; and Pseudo-Kodinos, Traité des Offices, ed. Verpeaux, J. (Paris 1976) 189–246 Google Scholar. On the secular, religious and popular holidays of the year, see Bréhier, L., Les institutions de l’Empire byzantin (Paris 1949) 82–83 Google Scholar, and Koukoulès, P., (Athens 1948-55) II, pt. 1, 7–38.Google Scholar

19. Unless there is some explicit statement to this effect in the sources, I see no way to confirm this hypothesis. There are 158 documents with precise dates of issuance (day and month), and while it certainly is possible to determine what percentage of these were issued on the kind of holidays De cerimoniis, etc., mention, the results would have little significance, because documents with precise dates are, by their nature, special, and it might be precisely because they were issued on holidays that their precise date was included. In any event, the question is really tangential to a study of seasonality. Since every month had a good share of holidays (with the exception of July), there were always a few occasions during any month to sign documents (assuming, that is, that documents tended to be signed on holidays). In other words, a month with five holidays need not necessarily have more documents than a month with two. If, however, a month had no holidays (as I suggest might be the case with July), the situation is somewhat different.

20. Oikonomidès, Listes, 215 and note 252.

21. De cerim.(Bonn ed.) 114ff. and 186–87; (Vogt ed.) 106ff. and 174. Grumei, V., Les regestes des actes du patriarcat de Constantinople, vol.1, Les actes des patriarches, fase. II (Paris 1936) no. 669.Google Scholar

22. One might add that annual fasts could possibly have had some effect on the seasonal index of chancery activity, if, for example, it was regarded as inappropriate for the emperor to go hunting during such a fast period. It is interesting that the standard fasts still leave July free (the Fast of the Apostles lasts from the day after All Saints’ Sunday to the eve [June 28] of the feast of SS. Peter and Paul, and the Fast of the Dormition lasts from August 1 to 14).

23. On this period, see, e.g., Ostrogorsky, G., History of the Byzantine State, rev. ed. (New Brunswick, New Jersey 1969) 401–402.Google Scholar

24. Cf. Oikonomidès, Chancellerie impériale, 168. On this ‘household’ government, see Angold, M., A Byzantine Government in Exile (London 1975) 3f.Google Scholar, and cf. Ostrogorsky, History, 427.

25. Oikonomidès, Chancellerie impériale, 182 note 75, and 193–94.

26. See Laiou, A., Constantinople and the Latins (Cambridge, Mass. 1972) 5, 115–16 Google Scholar, and 256 (on decentralization during Andronikos II’s reign), 120, 124, 194, and 277 (on the corruption and inefficiency of civil servants), and 187–88 (on the decision to withhold the salaries of the palace staff in 1304 and 1305).