War Of The (Gold) Worlds

In my previous War of the (Gold) Worlds article I presented a method for constructing three hypothetical price series from observed Gold prices. Each of the series reflect how the cash Gold price would have evolved if trading were always isolated to only one fixed intraday time period during the 24 hour trading day.

The method for calculating the hypothetical price series holds together fine. However, one of the cash Gold prices I used as an input to the calculations of the series was in error. I mistakenly assumed that 10 times the closing price of the SPDR Gold Trust ETF was a reasonable estimate for cash Gold prices when the NYSE closes. This article presents calculations of the hypothetical series where a more accurate method is used for estimating cash Gold prices when the NYSE closes.

The objective of these articles is to determine if there is any quantitative evidence that supports the perception that selling forces in the international Gold bullion trade have a greater tendency to appear on a regular basis during certain times of the 24 hour trading day. An examination of the corrected hypothetical price series suggests that there are indeed time periods of the day that sometimes do have a tendency towards buying and others that sometimes have a tendency towards selling.

An Incorrect Premise

Some observant readers of my previous War of the (Gold) Worlds article pointed out that one of my key assumptions was largely incorrect, (see reference [1]). I assumed that 10 times the closing price of the SPDR Gold Trust ETF was a reasonable estimate for the prevailing cash price of Gold when the NYSE closes. (Formerly called the StreetTracks Gold Trust and herein referred to as GLD for brevity).

While this was a good assumption when GLD first started trading in November 2004, the quantity 10 times GLD presently under estimates cash Gold prices by a significant amount in percentage units. I'll go into more detail in a moment, but the brief explanation for this is because the number of ounces of Gold per share of GLD has slowly decayed over time from its initial value of 1/10 ounce per share. There is nothing nefarious here. A recent GLD prospectus clearly states; "The Trust does not generate any income and as the Trust regularly sells gold to pay for its ongoing expenses, the amount of gold represented by each share has gradually declined over time."

This oversight introduced a serious error into the returns for sessions one and three that started small and gradually became worse as time passed. As a result, the hypothetical price series I presented for sessions one and three were wrong and should be ignored. (The series I presented for session two, or the "London" session were not influenced by this error and require no corrections of any kind).

The good news is that there is a straight forward fix for this problem. There is a variety of detailed data relating to the financial status of the ETF available from the SPDR Gold Trust website. There are two bits of information provided there that make it easy to get what I wanted all along; a sampling of the prevailing cash price of Gold when the NYSE closes. The resulting estimate is perhaps not perfect, but it is a pretty good one.

A handful of readers got the impression I was trying to belittle or otherwise berate the GLD ETF. The only motivation for using GLD closing prices in my analysis was that it appeared to be an easy way to get (estimates of) historical prices of cash Gold sampled at the time of day when the NYSE closes. In the course of researching how the ETF operates, I can say that SPDR Gold Shares does a good job of providing detailed financial records and in a timely fashion. I did not see any evidence that GLD operates in a manner contrary to that represented by the fund management.

Estimating Cash Gold at the NYSE Close

In this section I will provide the details of how some of the data available from the SPDR Gold Trust website can be used to calculate a good estimate of the prevailing cash price of Gold when the NYSE closes. If you don't care about these details you can just skip to the following section to see a plot of the corrected hypothetical time series for all three sessions.

The data that I used can be found on the "Historical Archive" page of the SPDR Gold Trust website, (see reference [2]). Recent data is displayed in tabular form and a complete historical record can also be downloaded in comma separated values format. When referring to any of this data I use the same terminology as used by the SPDR Gold Trust website to avoid any confusion.

The quantity of Gold that each share represents on any given day can be determined from the values of "NAV per GLD in Gold" that is reported in units of percent of 1/10 ounce. For example; the value reported on 14 November, 2008 was 98.42574% which translates to 0.09842574 ounces of Gold per share (their choice of units basically saves the printing of some extra leading zeros).

The other bit of information one needs is what is termed the "Indicative Price of GLD" which is labeled as being sampled at 4:00 PM New York time. Although it is not explicitly spelled out, this appears to be the amount of Gold per share valued at the prevailing price from the cash market. This should not be confused with the closing market price of GLD that I used in the previous article. (The closing price of GLD is also reported there and an inspection of the values make it clear that these are two entirely different quantities). I did not use the closing prices of GLD in this analysis.

To get an estimate of cash Gold prices at the NYSE close I simply divided the "Indicative Price of GLD" by the calculated quantity of Gold that each share represents on that day. I should point out that there are other ways to arrive at the desired estimate as the website provides quite a variety of historical data. Table 1 below presents the calculation of the 5 most recent estimates from the data.

As I noted above, this estimate may not be perfect. While certainly better than the quantity 10*GLD there is a small issue I noticed that should be mentioned for completeness. This issue might be considered minutia for some, they can skip to the next section.

The SPDR Gold Shares website does not spell out the time of day associated with the reported values of "NAV per GLD in Gold", (nor could I find clarification in any of the other documents they make available). It is possible the the values they report for the "Indicative Price of GLD" are based on a value of "NAV per GLD in Gold" that corresponds to a slightly different time of day. Even if this is the case the errors introduced should be fairly small.

An inspection of Table 1 above shows that the change in "NAV per GLD in Gold" from day to day is quite small, so any intraday change will be even less. The error introduced into the estimate of cash Gold should be less than the error introduced from intentionally using a value of "NAV per GLD in Gold" from the day before or the day after the one of interest. For each day I calculated a reference price estimate using values of "NAV per GLD in Gold" from the day of interest. I also calculated estimates using "NAV per GLD in Gold" values from the day before and the day after. The differences in these price estimates for any given day relative to the reference were generally less than +/- 0.01 US$ with a very small number exhibiting differences of up to +/- 0.05 US$. Errors of this magnitude are tolerable for the purposes of this article.

The Corrected Hypothetical Price Series

In this section I will present the hypothetical price series using the corrected estimates for cash Gold when the NYSE closes. The procedure for calculating the hypothetical series themselves from observed prices is explained in detail in my previous article and does not require any corrections, (see reference [1]).

Even those readers that are familiar with my original article may not recall the details of how I partitioned each 24 hour trading day into three non-overlapping trading sessions. For the benefit of all readers, here is a review of the precise definitions of the three trading sessions.

The corrected hypothetical prices series are shown below in Figure 1. As before, I started the calculation of all of the series on Monday 29 November, 2004 which is a few days after the SPDR Gold Trust ETF first began trading on the NYSE. Unfortunately, it is not possible to choose starting dates that go back in time before the ETF began trading. The calculations used data up to and including that for Friday 14 November, 2008 so the chart covers a duration just shy of four years.

In the absence of the errors introduced by using a bad estimate for the Gold price when the NYSE closes, the hypothetical prices series exhibit more believable numbers. A few comments are in order for each of the price series.

For session one, or the "Asian" session, the hypothetical prices actually declined a little bit from the start into August 2005. From there it exhibits a fairly steady rise that lasted until March 2008. Since that time the prices have been quite volatile, (as they are for each of the three series) and have declined a bit. Over the entire duration covered, this trading session has contributed an upward influence on prices after all the day to day fluctuations are netted out.

For session two, or the "London" session, the ending price of the hypothetical series is 306.66 US$ which is certainly less than the starting value. One can conclude from this observation that this session has introduced a net downward influence on Gold prices during the period covered. While it is a little difficult to see because of the differences in the scaling of the price axis, this series is identical to that presented in my previous article.

The results for session three, or the "New York" session, are a little surprising. The ending price of the hypothetical series is 869.07 US$, meaning that it had the largest net upward influence on prices of all three of the trading sessions. Prices made steady gains from the start into May 2006 and was followed by a sideways correction that lasted until roughly June 2007. From that point, prices rallied again into July 2008 and have since pulled back some.

Table 2 below shows the same statistics that I presented in the previous article, but using the corrected estimates for cash Gold when the NYSE closes.

Statistic	Session 1	Session 2	Session 3	NY EST
Starting Price (US$/OZ)	453.5449	453.5449	453.5449	453.5449
Ending Price (US$/OZ)	572.8873	306.6661	869.0660	742.2449
Total Return (raw)	1.2631	0.6762	1.9162	1.6365
Max Daily Return (%)	3.3948	4.6623	6.6595	11.3945
Mean Daily Return (%)	0.0243	-0.0407	0.0676	0.0512
Median Daily Return (%)	0.0195	-0.0319	0.0945	0.0597
Min Daily Return (%)	-3.8651	-4.4721	-6.1587	-8.5295
Number of Up Days (%)	52.1830	46.3617	57.2765	53.3264
Mean Return of Up Days (%)	0.5150	0.5922	0.5780	1.0070
Number of Down Days (%)	47.7131	51.4553	42.6195	46.5696
Mean Return of Down Days (%)	-0.5096	-0.6092	-0.6141	-1.0320

If you have a preference for looking a statistics versus graphs, Table 2 above tells a similar story. An examination of either the sample mean or median values of the daily returns shows that both the "Asian" and "New York" sessions had an upward influence on prices while the "London" session had a downward influence on prices (after netting out all the day to day fluctuations).

Summary

Many observers of the international Gold bullion trade have commented that when heavy selling pressure appears, it seems to happen more often than not during the hours when the New York markets are open for business. The objective of these articles was to determine if there was any evidence that supported that perception.

A technique was presented in my previous article for the calculation of three hypothetical price series, each one representing how the Gold price would evolve if trading occurred only during a certain period of the 24 hour trading day, (see reference [1]). This approach is not a prediction of any sort, it is just a different way of looking at the historical record.

This article presented a correction to the methodology used to estimate the prevailing price of cash Gold when the NYSE closes from data made available by the SPDR Gold Trust ETF. An inspection of the three hypothetical price series provides quantitative evidence that the intraday time period between the AM and PM London fixes has exerted a net downward influence on cash Gold prices over the last four years. The intraday time period from the PM London fix to the closing of the NYSE has actually had a net upward influence on prices over the last four years.

Future Work

One of the limitations of the analysis I have presented is that I was constrained to partition the 24 hour trading day into sessions that start and end at times when price quotes for cash Gold are readily available. To perform the analysis with trading sessions that have arbitrary start and end times requires price quotes for cash Gold that are sampled frequently, preferably at intervals of 30 minutes or less. Utilizing this sort of data also has the advantage that the introduction of a bias error that might arise from using disparate data sources is much less likely. (This is what occurred with my estimates made from 10 times the closing price of GLD).

Nick Laird, the proprietor of ShareLynx.com has kindly provided me with just this sort of data for my analysis, (see reference [3]). I am planning to publish a follow up to this article with trading sessions that start and end at times that better reflect the actual business hours of major global exchanges.

Footnotes, References

Appendix

In this appendix I have tabulated the first and the last 15 values of each of the hypothetical price series for the benefit of those who might want to reproduce the calculations or just confirm their understanding of them. (Had I tabulated the last 15 values for the previous article, I probably would have spotted the issue with 10*GLD!)

Date	AM Fix	PM Fix	NY EST	Session 1	Session 2	Session 3
2004-11-29	450.40	451.25	453.5449	453.5449	453.5449	453.5449
2004-11-30	452.00	453.40	451.4491	452.0000	454.9497	451.5934
2004-12-01	451.10	452.85	453.7444	451.6504	456.7146	452.4853
2004-12-02	454.35	454.20	449.5488	452.2532	456.5638	447.8517
2004-12-03	449.15	448.65	456.3453	451.8520	456.0556	455.5333
2004-12-06	455.75	453.05	452.0491	451.2626	453.3538	454.5269
2004-12-07	453.75	451.80	450.8438	452.9606	451.4055	453.5649
2004-12-08	445.75	436.90	440.1482	447.8429	442.4432	456.9369
2004-12-09	437.35	437.10	437.7424	444.9958	442.1903	457.6085
2004-12-10	433.90	434.00	434.5469	441.0897	442.2922	458.1851
2004-12-13	436.95	435.10	439.0425	443.5290	440.4196	462.3368
2004-12-14	438.60	437.10	435.5959	443.0820	438.9133	460.7459
2004-12-15	437.80	439.00	440.0916	445.3240	440.1164	461.8915
2004-12-16	441.25	439.50	437.1455	446.4962	438.3709	459.4171
2004-12-17	439.20	438.90	441.5417	448.5946	438.0714	462.1823
...						...
2008-10-27	720.50	730.50	731.0458	545.0736	306.1001	901.2941
2008-10-28	745.00	730.50	745.9455	555.4779	300.1424	920.3508
2008-10-29	749.25	764.00	756.7403	557.9387	306.0511	911.6055
2008-10-30	772.25	755.25	737.9489	569.3738	299.3139	890.7227
2008-10-31	728.50	730.75	724.5466	562.0834	300.2383	883.1612
2008-11-03	734.00	729.50	723.5449	569.4171	298.3976	875.9517
2008-11-04	734.00	741.25	765.9406	577.6451	301.3450	905.1291
2008-11-05	753.25	753.75	740.6483	568.0744	301.5450	889.3961
2008-11-06	739.00	754.50	734.7436	566.8102	307.8697	866.1074
2008-11-07	742.00	735.25	737.3404	572.4080	305.0690	868.5698
2008-11-10	751.75	753.00	746.5404	583.5944	305.5763	861.1188
2008-11-11	741.75	733.75	733.2488	579.8496	302.2805	860.5306
2008-11-12	731.50	724.75	711.7430	578.4667	299.4912	845.0868
2008-11-13	714.00	713.50	738.9404	580.3010	299.2815	875.2190
2008-11-14	729.50	747.50	742.2449	572.8873	306.6661	869.0660