Full index of posts »
StockTalks

AMZN CFO Resigns $AMZN http://seekingalpha.com/p/1xe43 Sep 6, 2014

Aspen Shareholders Reject Endurance's Acquisition Bid $AHL http://seekingalpha.com/a/1ebyz Jul 29, 2014

Three Reasons Aspen Insurance Should Accept The Buyout From Endurance $AHL http://seekingalpha.com/a/1a2pp Apr 19, 2014
Latest Comments
 Brian Abbott on Selling Puts On The Dogs Of The Dow >>But why do you choose the smallest pric...
 JohnMartinez on Selling Puts On The Dogs Of The Dow Brian, thanks for the article. I learned some. ...
 Brian Abbott on Discounting Everything By 30% Micheal thanks for commenting.Another point of...
 Michael Nau on Discounting Everything By 30% I agree that psychological factors often cause ...
Most Commented
 Discounting Everything By 30% ( Comments)
 Selling Puts On The Dogs Of The Dow ( Comments)
Posts by Themes
alpharichinvestmentstyle,
alpharichopportunity,
asset allocation,
biases,
biotechnology,
bonds,
brianabbott,
cognitive bias,
commodities,
correlation coefficients,
etfanalysis,
eventspecialsituations,
financial,
GE,
goldandpreciousmetals,
hedge funds,
incomeinvestingstrategy,
INTC,
investingideas,
longideas,
Longterm investing,
MLP,
options,
Ownership mentality,
pension funds,
PFE,
portfoliostrategyassetallocation,
private equity,
propertycasualtyinsurance,
quickpickslists,
REIT,
research,
saexclusive,
smallcapinsight,
T,
transformativeacquisition,
us
Are Commodities A Leading Or Lagging Predictor Of Equity Prices? 0 comments
(This analysis was performed in 2006)
EXECUTIVE SUMMARY
There have been many attempts to accurately forecast the direction of public equity markets. Because of the number of participants in these markets, any true predictive relationships tend to be extremely shortlived and selfcorrecting, through a process known as arbitrage in financial parlance. The Dow Jones Industrial Average of 30 leading industrial companies is one of the most widely followed equity indices in the world. Commodity prices are an important factor in the cost of industrial production and thus are expected to affect the profitability of industrial companies, and possibly the valuation of industrial companies in the equity markets. Other factors that may be associated with the economy, such as Consumer Price Index (NYSEARCA:CPI) and the National Bureau of Economic Research (NBER) economic cycles, could also affect the profitability and equity valuation of industrial companies.
This analysis analyzed potential associations with the DJIA and the Commodity Producer Price Index (CommodityPPI). Initial analysis revealed a significant but relatively weak positive linear association between the CommodityPPI and the next month's DJIA performance. Additional variables that were included in further modeling included CPI, NBER economic cycle periods, and pre/post1970 era analysis. Inclusion of the NBER economic cycle in the model strengthened the predictive power of the CommodityPPI and next month DJIA by identifying that this relationship existed only during months of economic contractions (from peak to trough of the economic cycle). Further analysis revealed that this relationship existed only in the months prior to 1970 and not subsequently. A potentially confounding effect was identified between CPI and CommodityPPI.
To account for all of these factors, a complex multiple regression model was constructed including all of the variables mentioned above. The strongest association in this model was between the CommodityPPI and DJIA. However, the association between CommodityPPI and CPI could not be excluded as a confounding factor. Another complicating factor was that a fair number of outlier extreme values in the DJIA could not be accounted for in the multiple regression model. This was felt to be an artifact of kurtosis, a phenomenon previously described in equity market performance with the occurrence of much larger events than would be anticipated statistically.
Potential areas for future study are discussed, including the use of a more modern equity market index of commodityintensive industrial companies, since changes in the DJIA component companies could explain the fact that the strong correlation between CommodityPPI and DJIA during contracting economies appeared to breakdown sometime subsequent to 1970. Also, better indicators of inflation could be included, such as M2 money supply and interest rates, to help offset the potentially confounding influence of monetary inflation on the variables in this analysis
INTRODUCTION
Background
The economy in capitalist societies is typified by boom and bust cycles of various periodicities. Despite intense study, few experts have been able to predict the secular trends of growth and contraction. Economists have proven to be notoriously unreliable at predicting the onset of economic recessions accurately, for example.1
Thus, it is not surprising that predictions of equity market performance are even worse in accuracy. A successful model of macroeconomic leading indicators might be able to assist portfolio managers in repositioning investment portfolios to avoid losses and to take advantage of gains in different asset classes.
The Macroeconomic Business Cycle
Recessions are described as period of economic contraction, in which previous malinvestments are corrected through revaluation of asset prices and even bankruptcies. Expansion cycles are characterized by economic growth and increased borrowing for buildout of capacity. Historical data of "official" business cycles are collected by the National Bureau of Economic Research (NBER).2
Macroeconomic input factors of production are tracked in order to explain the shortterm economic cycles of expansion and contraction. These inputs consist of labor, borrowed capital, and raw materials. Economic indicators measuring these input factors are divided into coincident, leading, and lagging indicators, depending on how the number peaks are in or out of phase with the economic expansion/contraction cycle.3 Coincident indicators include payroll data. Leading indicators include average work hours, durable goods orders, M2 money supply, new housing starts, and consumer expectations.4
The stock market itself has sometimes been viewed as a leading indicator for the economy.5,6 Examples of lagging indicators include unemployment rate and changes in the consumer price index.
One difficulty of interpreting leading indicators is that they are by definition out of phase with the current economy. It is not uncommon for many business leaders and investors alike to ignore or misread leading indicators. For example, businesses suffering through a recession are often reluctant to reposition for growth when leading indicators show an upcoming turnaround. Similar things are observed at the peaks of economic cycles, when businesses and investors may inappropriately extrapolate the recent wonderful growth in profits for years into the future. An awareness of these flawed emotional tendencies may allow astute investors to profit at important inflection points in the economic cycle.
The equity market as a proxy for macroeconomic performance
For the purposes of this analysis, I am attempting to use the stock market index returns of industrial companies as proxies for macroeconomic performance. Obviously, stock markets may not reflect macroeconomic output over the short run, which I will arbitrarily define as 2 years or less. This is likely due to inherent biases among investors and corporate management, including overconfidence and inappropriate extrapolation of recent news.7 However, over the long term (10 years or more) such performance should be highly correlated with macroeconomic output, given the size of the United States public equity markets. If it were otherwise, then the stock market could be severely mispriced and result in large potential opportunities for arbitrage8 to attempt to bring valuations back in line. Basically, this is a major premise of the efficient market hypothesis and of capitalism theory itself  that any short lived price anomaly serves as an opportunity for an astute investor, and will be corrected in the long run.9
In summary, the capital markets can be irrational or inefficient over short periods of time, but over longer periods the performance should be driven by fundamental value rather than the whims of investor emotions. Since commodity prices are one of the major determinants of the cost of production of industrial companies (such as those within the Dow Jones Industrial Average), commodity prices may provide important signals about the macroeconomy. Because the public equity markets represent the overwhelming majority of value of United States firms, the fate of equities should reflect the overall fate of the macroeconomy.
Definition of the problem
This analysis will test whether the commodity producer price index has correlations with equity performance, and if so, whether they are a leading or lagging indicator.
DATA
Data were collected from a variety of public sources (see variable table and references). I personally downloaded each data series from publicly available websites. Monthly data series for a variety of indicators were obtained and collated into a single spreadsheet file. Data were verified for completeness and proper data transfer. After collection, data were transferred to a JMP data sheet. After verifying accurate transfer of data, additional data columns were generated for the derived and calculated fields, as described below.
Another categorical data series was added to indicate the NBERdefined months of economic contraction (from peak to trough of the economic cycle) and expansion (trough to peak); there were 21 complete cycles from 19902006, and we are currently still in the midst of an expansionary phase at the time of this writing in December 2006 (although some leading indicators appear to have trended downward).
Variables
The proposed analysis will use multiple regression of monthly time series data. The primary response variable will be the Dow Jones Industrial Average (DJI) monthly % change, which is a continuous variable. This was derived from the DJI data series, which is a continuous metric variable with units of nominal US dollars.
The predictor variables in the regression analyses will consist of the following:
Table 1. Variables
Monthly Data series
Date range
Source Reference
Response variable:
DJIA, real (monthend closing prices)
19002006
11
Predictor variables:
CPI, nominal
19132006
12
CPI, monthly % change
(calculated)
(calculated)
CommodityPPI, monthly % changes
19132006
13
NBER Economic Contraction (0=no, 1=yes)
19002006
Limitations of the data
For the stock market indices, the foremost weakness of analyzing longterm data is that the stock indices have changed over the years. Reasons for changes in the index components can be due to mergers or bankruptcies. Since bankrupt companies have been removed from the indices and replaced with healthier ones, this introduces a survivorship bias and may overstate long term equity market returns; this is inherent in any long term equity market index analysis. This is a limitation of all available indices and there is effectively no way to avoid it.
Another potential confounding factor for the predictor and response variables is the influence of inflation. While beyond the scope of discussion, inflation is generally a monetary phenomenon, whereby increasing amount of money are placed into circulation through the fiat currency process of fractional reserve lending. In order to offset the longterm trend for fiat currencies to lose purchasing power through inflation, the prices of assets generally rise over time. This would be expected to cause concurrent increases in CPI, PPI, and Dow Jones. Even if inflation is an underlying factor behind increases in CommodityPPI and the DJIA, it is possible that a slight investment advantage could be gained if the optimum leading and lagging time period could be identified.
Another limitation of the data is that equity index returns have been described as having kurtosis.14 This phenomenon occurs when a variable exhibits a form of nonnormal distribution, resulting in "fat tails". In other words, although the majority of time periods appear to follow a normal distribution, extreme values (<5 and >95th percentile) occur at a much high frequency than would be predicted by a normal distribution. The issue of market kurtosis was the subject of an entire recent popular book.15 Kurtosis is a difficult problem to deal with in financial market analysis, and has been the cause of demise of multiple hedge funds that initially exhibited strong performance until becoming overwhelmed by statistically rare events that resulted in much larger losses than predicted by the computed model.16,17
Finally, the data are limited by a lack of granularity. The statistics that are publicly released are often limited to the tenth of a percent. This results in a large cluster of events between 0.5 and +0.5%, with a number of outliers. While transformation may help result in a more normal distribution, it may also amplify the limited granularity of the data. A related issue is that the producer price index often reports a zero percent change in many months, since many months contain rather small changes in prices that round to zero, potentially hiding tiny directional movements in prices.
METHODS
First, each of these data series were analyzed for normal distributions using summary statistics, histograms, and normal quantile plots. Nonnormal data were considered for transformation when appropriate. The time frames used were March, 1926 through October, 2006, since this was the time frame for which data for all variables were available.
Response variable, summary statistics
As shown in Figure 1, the monthly percentage changes of the Dow Jones Industrial Average (DJIA) do not follow a normal distribution. The distribution of monthly changes shows a sharp peak, but with an increased number of events outside the two standard deviation bands. Further analysis reveals a heavy clustering of values within 1% and +1%. The mode of +0.3% is approximately equal to the mean.
As expected, the DJIA monthly changes exhibit kurtosis, or fat tails. For examples, the negative 23.2% change for the month of October 1987, is 4.3 standard deviations from the median monthly change. A change of this magnitude should occur less than 0.00007% of months. However, October 1987 is not even the largest drop. The largest monthly drop during the analyzed time period occurred during September 1931 (30.7%, 5.7 standard deviations). A change of this magnitude is expected to occur less than 0.00000006% of months, if returns followed a statistically normal distribution. Thus, during the 968 months under observation, 0.000058 months of this magnitude of a monthly change in DJIA would have been expected had the results followed a normal distribution. Thus, a data set of over 206,000 years would be needed!!
Several procedures were attempted to produce a more normal distribution, in order to facilitate further analysis. Other transformations that were attempted include: natural logarithm, base 10 logarithmic, exponential transformation, reciprocal, reciprocal of the natural logarithm, square root, and quartic root of DJIA. The quartic root of DJIA appeared to produce the most normal distribution with the smallest number of outlier events (Figure 2). However, the use of the quartic root would have obscured the directional sign of the monthly changes in DJIA, which is the primary focus of the analysis. Thus, untransformed DJIA data were used for subsequent analysis.
Figure 1. Dow Jones Industrials, monthly changes (%): Summary Statistics
(click to enlarge)
Figure 2. Transformation: Quartic Root(DJIA)
(click to enlarge)
Predictor variable, summary statistics
The CommodityPPI variable is the primary predictor under consideration in this analysis. The raw data of monthly percentage changes in CommodityPPI are shown in Figure 3. The data do not have a normal distribution, as shown in the normal quantile plot. Among the transformations tested, the natural logarithm produced a slightly more normal distribution as judged by the normal quantile plot. The histogram appears to be both rightward skewed with outliers and also with an exaggerated peak to the left of the mean. Thus, subsequent analysis used the raw, untransformed data for commodityPPI. The rationale for this is that part of the nonnormality of the data was due to an increased number of months with zero percent reported change. There should be no directional bias from these zerochange months when comparing to the DJIA monthly changes, so the statistical model under development would be expected to be underpowered to detect a meaningful correlation (type one error). This was felt to be preferable to the risk of introducing a type 2 error (in describing a nonexistent correlation), which might have occurred if excessive transformation of nongranular data were performed.
Figure 3. CommodityPPI, monthly % change, Summary Statistics
(click to enlarge)
Linear regression analysis of CommodityPPI and next month DJIA.
The first testing focused on comparing the commodityPPI monthly changes (predictor variable) with DJIA monthly changes (response variable). Untransformed data series were used for each variable. When same month data were used, no significant relationship was identified by linear regression (correlation coefficient=0.003, p =0.081), as shown in Figure 4.
The commodity producer price index (PPI) was lagged for 1, 3, and 6 months, and also advanced by 1, 2, 3, and 6 months relative to DJIA. The commodity PPI is composed of a basket of common commodities. Because the hypothesis was that CommodityPPI might serve as a leading or lagging predictor of DJIA performance, various leading and lagging periods for CommodityPPI were tested. Linear regression statistics from these analyses are summarized in Table 2. To summarize these findings, the strongest correlation coefficient and statistical significance were identified when the CommodityPPI had a lead time of one month compared to the DJIA. With a one month lead time, the linear regression analysis identified a correlation coefficient of 0.011, with p=0.0009.
Table 2. Linear regression statistics with various leading and lagging time period of CommodityPPI, when compared to DJIA monthly changes.
CommodityPPI
rsquared
slope
P value
Lag 6 months
0.000144
0.00402
P=0.7106
Lag 3 months
0.000031
0.00039
P=0.8631
Lag 1 months
0.000095
0.00067
P=0.7628
Same month
0.003
0.00388
P=0.0812
Lead 1 months
0.011
0.00739
P=0.0009
Lead 2 months
0.00456
0.00469
P=0.036
Lead 3 months
0.00455
0.00467
P=0.0364
Lead 6 months
0.000037
0.00042
P=0.8500
Figure 4. DJIA by CommodityPPI, one month leading
Further analysis was performed to see if different relationships existed during expansionary, contractionary, or both types of months of the NBER economic cycles. The summary statistics for the monthly changes of CommodityPPI by type of NBER month are shown in Figure 5. Nonparametric tests (Wilcoxon, Median, and Van der Waerden tests) showed significant differences in the CommodityPPI during contractionary months, which were lower in magnitude, compared to expansionary months (data not shown). Linear regression slope statistics for these analyses are summarized in Table 3. During expansionary economic phases, there is no significant relationship with CommodityPPI and next month DJIA returns. However, during contractionary months there appears to be an association of CommodityPPI with DJIA (Figure 6). The slope is higher on this linear regression analysis, as is the correlation coefficient, compared to the expansionary months and the nonsegregated data CommodityPPI data set. This suggested that the association of CommodityPPI and next month DJIA (identified above) is primarily limited to the contractionary months of the economic cycles.
Table 3. Linear regression statistics with CommodityPPI and nextmonth DJIA monthly changes, according to NBER economic cycle phase.
CommodityPPI
NBER cycle
rsquared
slope
P value
Lead 1 month
All months
0.0114
0.00739
P=0.0009
Lead 1 month
Contractionary months
0.0384
0.0203
P=0.0034
Lead 1 month
Expansionary months
0.00006
0.00154
P=0.44
Figure 5. CommodityPPI monthly changes, by NBER economic cycle phase.
(click to enlarge)
Figure 6. DJIA by CommodityPPI during contractionary economic cycles
Consumer Price Index monthly changes
The Consumer Price Index (CPI) was analyzed due to the likely confounding effects of consumer inflation over the long time period being analyzed. As with other data sets, the summary statistics for CPI showed a slight kurtosis, although the means and median values are similar, the histogram shows no perceptible skewing, and the data points nearly follow a normal distribution on the normal quantile plot (data not shown). Thus, nontransformed data were used with nonparametric tests for statistical comparisons with other data series.
Preliminary analysis for underlying influence of consumerled inflation, as measured by the monthly percentage changes in the CPI, was performed. A number of leading and lagging time frames (CPI lagged by +6, +3, +2, +1, 0, 1, 2, 3, and 6 months) were tested. A two month leading time from for CPI relative to DJIA resulted in the only statistically significant positive association on linear regression analysis, as shown in Figure 7. Despite the statistical significance of this relationship (p=0.0048), the relationship is not of large magnitude, with a slope of 0.0016 and a correlation coefficient of 0.008. The other leading and lagging time periods tested showed no significant relationship at all between DJIA and CPI, which further indicating that there was not any pervasive association between the two measurements.
Figure 7. DJIA by CPI (leading, two months)
To assess for a confounding relationship between CPI and CommodityPPI, a linear regression analysis was performed on these two variables. As shown in Figure 8, there was a positive linear relationship between these two variables. Also, note that the time frames with the strongest association with DJIA were used in this and subsequent analyses (CommodityPPI leading one month, and CPI leading 2 months)
Figure 8, Linear regression of CommodityPPI (leading one month) and CPI (leading two months)
Because NBER economic cycles were found to affect the strength of the linear relationship between DJIA and CommodityPPI, further analysis was performed. Figures 9a and 9b show the results of linear regression between CPI and CommodityPPI in expansionary and contractionary months, respectively. While significant positive linear correlations were found in each series, the slope was higher in the contractionary months (1.03) compared to expansionary months (0.65). This indicated that the confounding relationship between CPI and CommodityPPI is sensitive to the economic cycle and is most profound during times of economic slowing.
Figure 9a. CommodityPPI by CPI during expansionary economic cycles
Figure 9b.CommodityPPI by CPI during contractionary economic cycles
Performance during expansionary and contractionary economic cycles
Evaluation of DJIA monthly performance according to NBER economic phase revealed no significant differences using nonparametric testing (data not shown). However, closer analysis revealed the standard deviation of monthly changes during months of a contractionary economy (0.108%) was twice that of expansionary months (0.057%). The fact that equity returns were not significantly lower during a contracting economy at first appears to be counterintuitive, but actually is consistent with many claims that the equity markets often anticipate the subsequent performance of the economy. This is also consistent with the commonlyheld belief that equity markets are more volatile during a less favorable economy.
Similar analyses were performed for the CPI and commodityPPI monthly changes. Unlike DJIA, both variables were found to be significantly lower during months labeled by NBER as contractionary, as shown in Figures 10a and 10b. These results are as expected, as consumer and producer price inflations are generally lower during contractionary months. Also, the standard deviations for both variables were slightly higher during the contractionary months (although less than 20% higher, in contrast to the DJIA which had twice the standard deviation during these economic periods). Thus, a contracting economy resulted in lower monthly changes in CPI and CommodityPPI but not DJIA. However, the opposite was true in terms of volatility, as measured by standard deviation, which was much higher in DJIA during a contracting economy, while being nearly unchanged for CPI and CommodityPPI.
Figure 10a. Analysis of commodityPPI monthly changes according to NBER economic cycle.
Figure 10b. Analysis of CPI monthly changes according to NBER economic cycle.
Subsequent analysis was performed to determine whether the significant relationship between commodityPPI and DJIA was present during pre and post1970 era. (Note that this time point was arbitrarily chosen.) These results indicated a significant association between DJIA and CommodityPPI was present only during the pre1970 era, with no relationship at all in the post1970 era (data not shown). To incorporate this finding into the model under development, the effects of CommodityPPI on nextmonth DJIA were analyzed by linear regression according to both NBER economic cycle and pre vs post1970 era. In this analysis, only the pre1970 contractionary months revealed any significant relationship (figure 11). In these months, there was a significant positive relationship between CommodityPPI and nextmonth DJIA monthly changes. The slope for this linear regression was 4.3% with a correlation coefficient of 0.21. This compares favorably with the earlier linear regression analysis (Table 2) and suggests that by narrowing down the time frame to pre1970 contractionary economic months, a stronger association between commodity prices and DJIA performance has been identified.
Figure 11. Linear regression analysis for DJIA according to CommodityPPI (leading one month), for contractionary economic months prior to 1970.
A multiple regression model was constructed to combine the variables discussed above. In this model, the DJIA was the response variable. The predictor variables included the continuous data series, ConsumerPPI and CPI, since both variables were shown to have relationships with the DJIA. NBER contraction and pre1970 were used as categorical predictor variables, since they had been shown to identify time periods during which the other variables had a stronger relationship with DJIA. The results of leverage plots for each variable are shown in Figure 12a. The multivariate model statistics are shown in Figure 12b. It is important to note that multicollinearity is present due to the presence of the identified strong relationship between CommodityPPI and CPI, and future analyses would require a better accounting for the potentially confounding effects of inflation among all of these variables. While the CPI has a linear relationship with ConsumerPPI, it is not a statistically significant predictor of DJIA within this multiple regression model. The only significant independent predictors of DJIA performance in this model appear to be CommodityPPI (leading one month) and NBER contraction months, as indicated with p values of 0.0218 and 0.0314 respectively.
Residual analysis from this multiple regression model reveals a somewhat sigmoidalshaped normal quantile data plot with a few outliers outside the normal distribution, as shown in figure 12c. Figure 12d shows a time series model of the multiple regression residuals that reveals no apparent trends in residuals, but occasional large outliers appear to be scattered randomly throughout the data series. Figure 12e shows a residual versus predicted plot, which reveals no apparent curvilinearity but a possibility heteroskedasticity, with a higher density of points in the central portion of the plot. This could represent an effect of the outlying fat tails, represented by the single dots, which appear far outside the very dense portion where most points are found. This could be an artifact of the kurtosis in many of the data series that was discussed earlier. Figure 12f shows the histogram of Cook's D analysis of residuals from this model, which shows a number of influential outliers. Again, this is consistent with the hypothesis that kurtosis of the variables produces outlier months with exceptionally large monthly changes that are difficult to account for within a multiple regression model. Finally, figure 12g shows studentized residuals and hat analysis for the multiple regression model. The studentized residuals show a fair number of outliers, reflecting the kurtosis mentioned above. In summary, residual analysis reveals a possible influential effect of some of the extreme value outliers. While removal of the outliers would likely improve the correlations in the multiple regression model, it would also render it useless as a predictive tool.
Figure 12a. Leverage plots of predictor variables in multiple regression model of DJIA
(click to enlarge)
Figure 12b. Multiple regression model for DJIA, by ConsumerPPI, CPI, NBER economic phase, and pre/post1970 eras.
Figure 12c. Residual analysis for the multiple regression model.
(click to enlarge)
Figure 12d. Time series plot of residuals from the multiple regression model.
Figure12e. Residual by predicted plot for the multiple regression model.
Figure 12f. Histogram of Cook's D Influence of Multiple Regression model residuals.
Figure 12g. Studentized residuals and hat values of multiple regression model residuals.
(click to enlarge)
DISCUSSION AND CONCLUSIONS
A positive linear relationship was identified between CommodityPPI and DJIA from 1926 and 2006. While statistically significant, it was relatively weak in terms of magnitude, as measured by slope and correlation coefficient on linear regression analysis. By analyzing multiple factors, a stronger relationship was uncovered between these variables for pre1970 months during periods of economic contraction.
A multiple regression model was constructed to test whether the factors shown to be individually important factors of DJIA performance could be combined in a more complex model. The residual analysis of the multiple regression model indicates that a significant effect of outliers creates difficulties in using this regression model for predictions. While removal of outliers would very likely strengthen the correlation coefficient, it would also render the model useless as a predictive tool. The effects of kurtosis have created multibillion dollar losses for highly leveraged hedge funds that ignored the outlying fat tail events in their statistical models.
Multicollinearity is a difficult problem to deal with in this analysis. Because inflation raises the prices of essentially all assets (including equities), inflation likely remains a major lurking variable that influences DJIA, CommodityPPI, and CPI. Thus, subsequent analysis could include other measurements of inflation, such as interest rates and M2 money supply in an attempt to normalize for the effects of inflation.
These results highlight the complexity of attempting to predict equity market performance. Given the large number of professionals working to identify value with low risk, or "alpha" as described by the hedge fund industry, it is not surprising that relatively simple comparison of commodity price charts would fail to identify major correlations, or that correlations may be present in only certain time periods. Market experts might argue that the large number of mutual and hedge funds serve to immediately arbitrage any significant price anomalies. In fact, that any significant relationship at all was observed may be viewed as slightly remarkable, and further work could be performed to see if a modern relationship still exists between CommodityPPI and the performance of equity index averages.
Unfortunately for the initial hypothesis, the predictive value of CommodityPPI for nextmonth DJIA performance appears to have diminished since 1970. Several reasons could explain this apparent change in a previous significant relationship. One major reason could be a change in the types of companies included in the Dow Jones Industrials index. Modern companies that have been added to this index include the addition of more technology companies, including software companies (Microsoft), computer component companies (Hewlett Packard, Intel), and telecommunications (Verizon). These types of companies are expected to be less dependent on commodity prices and more dependent on intellectual property and brand equity. While these types of companies were added to DJIA to better reflect the composition of American production, it has also resulted in a shift away from commodityintensive industries toward softer, knowledgebased hightech industries.
Another interesting finding is the increase in volatility of DJIA monthly performance during a contracting economy. Our models do not account for an effect of volatility. However, many modern financial derivatives such as futures, options, and swaps do include volatility as a component of price. The same is true for many value at risk (VaR) calculations. This represents another potential avenue of investigation, since the increase in volatility could cause an increase in market instability or a market meltdown even in the absence of price changes, due to the way volatility is priced as a risk factor by many derivative models.
Further analysis could concentrate on identifying the best cutoff time frame, or on analyzing individual commodity prices to see which, if any, individual commodities might be most associated with DJIA performance. However, the observation that the association is significant only prior to 1970 suggests that little if any predictive role is likely to be found to aid in portfolio management decisions. If the diminution of predictive value of CommodityPPI was due to changes in the DJIA index, however, alternative indices of pure "hard manufacturing" companies could be analyzed to see whether CommodityPPI still has predictive value for these commodityconsuming industries.
REFERENCES
1findarticles.com/p/articles/mi_m1038/is_.../ai_12281225
2http://www.nber.org/cycles.html (accessed November 10, 2006)
3http://en.wikipedia.org/wiki/Leading_indicators (accessed November 9, 2006)
4http://www.conferenceboard.org/economics/bci/pressRelease_output.cfm?cid=1 (accessed November 9, 2006)
5www.econ.ilstu.edu/uauje/PDF's/issue1996....pdf
6economics.about.com/cs/businesscycles/a/....htm
7www.nber.org/digest/may05/w10813.html
8www.unc.edu/depts/econ/byrns_web/Economi....htm
9www.efalken.com/inefficient_markets.htm
10 http://www.nber.org/cycles/november2001/ (accessed December 12, 2006)
11http://www.djindexes.com/mdsidx/index.cfm?event=showHome&fmg=s (Accessed October 18, 2006)
12www.bls.gov/cpi/home.htm(Accessed November 3, 2006)
13http://www.bls.gov/ppi/home.htm (Accessed November 3, 2006)
14en.wikipedia.org/wiki/Kurtosis
15 Mandelbrot B, Hudson RL. The (Mis)Behavior of Markets: A Fractal View of Risk, Ruin, and Reward; Basic Books, 2004; ISBN 0465043550
16en.wikipedia.org/wiki/LongTerm_Capital_Management
17 finance.yahoo.com/columnist/article/futu.../10116
Disclosure: I am long SPY.
Instablogs are blogs which are instantly set up and networked within the Seeking Alpha community. Instablog posts are not selected, edited or screened by Seeking Alpha editors, in contrast to contributors' articles.
Share this Instablog with a colleague