"Bringing the measurements of critical economic activities into the twenty-first century by mining tracking data for an understanding of what American consumers were doing yesterday."
Leading Indicators are measures of (or statistics about) specific activities within the economy that generally precede or predict the general health of the economy as a whole.
The concept of an indicator that leads the performance of the economy has been around at least since Charles Dow published a series of editorials in 'The Wall Street Journal'. After Dow's death in 1902 the principles outlined in the editorials were recast by his successors at 'The Wall Street Journal' as the Dow Theory. Although Dow never referred to his ideas as 'the Dow Theory', he did propose that a bull market was likely to follow rallies in both his Industrial and Transportation Indexes. He further thought it likely that his Transportation Index would 'lead' the Industrial Index, since the railroads would have to carry goods before they could reach their ultimate markets.
In 1937 the U.S. National Bureau of Economic Research under the direction of Wesley Clair Mitchell produced a list of leading indicators in response to a concern about the status of the recovery from the 1937-1938 recession. The 1937 list of indicators began a compilation of nearly 500 time series of economic data, but did not yet include the concept of a single index (which was later developed by Henry Ludwell Moore).
Currently The Conference Board, a non-governmental organization, publishes the most closely watched set of indexes. Their Leading Economic Index is comprised of ten separate statistics: the U.S. economy's Money Supply (M2), the interest rate spread between long term and short term government bonds (the yield curve), an index of consumer expectations, the number of building permits issued, stock prices (using the S&P 500 stock index), supplier deliveries, average weekly manufacturing hours, average weekly initial claims for unemployment insurance, manufacturers' new orders for nondefense capital goods, and manufacturers' new orders for consumer goods and materials. Most of these statistics are updated by their government sources on a monthly basis.
Leading Indexes that rely on data published by governmental departments are generally updated monthly several weeks after the month's end. Often the governmental data includes some estimates and is necessarily preliminary, so a final set of numbers is published yet a month later. The resulting many week 'lag' in the most widely followed 'leading' index caused The Conference Board in January 2001 to revise the methodologies used in constructing their Index of Leading Indicators to make the index both more timely and more useful (see report by Robert H. McGuckin). Yet even with these revisions, the data used is (on average) still more than a month old at the time of publication, and the single monthly average value provided by each update results in only 12 measurements of any given index in any given year.
The Conference Board's Index of Leading Indicators has also traditionally heavily weighted monetary measures as predictors of future economic growth. These measures are manipulated by the Federal Reserve in efforts to stimulate the economy. During periods of expanding credit those measures have been very good leading indicators. But during times of credit contraction those measures have been far less predictive of future economic growth.
Additionally, most published 'Leading Indicators' use the value of stock market indexes as one of their key components. This presumes that the stock market itself is a predictor of the health of the economy, when arguably the relationship between the health of the equity markets and the health of the economy is both highly complex and subject to mutual feedback. The 2008-2009 recession provided some evidence of this connection, when the consequences of financial market problems with sub-prime mortgages significantly exacerbated the broader economic slowdown. And since most traditional 'Leading Indicators' already use recent historical stock market indexes as a component, they cannot have predictive value for those very same equity markets.
The 'Consumer Leading Indicators' on this site are much more timely than most other leading indicators, and they tightly focus on the U.S. consumer, which is the driving force behind 70% of the U.S. economy's activity. The increased timeliness is the result of two major improvements over other leading indicators:
1) We have taken Charles Dow's ideas to heart and we have moved as far 'up-stream' economically as possible - to the point where a consumer is actually making the initial purchase decisions for major durable goods. Our information is captured in some cases while the transaction is still being processed - before the retailer (let alone the wholesaler or manufacturer) is fully aware of the cash flows being generated.
2) We capture that data daily and publish the day-to-day results several times per week, unlike the monthly publication of monthly numbers typical of most other leading indicators. Additionally, we publish daily indexes for a number of separate sectors of the U.S. economy (e.g., the Automotive Index), and still more weekly sub-indexes of selected segments within those sectors (e.g., Domestic Autos or Luxury Autos).
3) We also differ from other indicators because our focus is exclusively on major discretionary spending of the U.S. consumer. This is the largest and most volatile portion of the U.S. economy, and the initiating force behind growth and contraction cycles. Other leading indicators heavily weight manufacturing data into their 'leading' indicators - activities which, from our perspective, are months or quarters 'down-stream'.
The Consumer Leading Indicators track consumer interest in major discretionary purchases. These typically include such items as automobiles, housing, vacations, durable household goods and investments. Not included would be expenditures that are more or less automatic and/or non-discretionary, such as groceries, fuel or utilities.
The index values are relative to the same period one year ago. An index value of 100 on any date indicates consumers showing the same level of interest in purchases as on that same date one year earlier. Thus an index of 105 indicates an increase in consumer interest of 5% relative to the same date a year earlier. An index of 95 would similarly indicate a 5% decrease in interest relative to the same period in the prior year.
There are a number of reasons why we have chosen to calculate the indexes this way:
1) Increases or decreases in consumer interest over the past 12 months are obvious from the index value itself, without the need to research through historic tables to find the values for prior dates. Thus an index value of 100 means that consumer interest is at the same level experienced a year ago; values above 100 indicate an upward trend while values below 100 show a declining level of consumer interest relative to the same date in the prior year.
2) By using this methodology each index is constantly being re-normalized. Unlike the consumer price index, for example, there is no set time period in the distant past when the index had some fixed value. Each index value therefore always reflects a '100 base' re-normalization relative to consumer interest levels 12 months earlier.
3) The economic implications of consumer behavior is markedly different in each sector of the consumer economy. Interest in automotive or housing purchases carry much greater impact on the overall economy than purchases of clothing or household goods. Yet within each sector it remains important to track sector-specific consumer interest in purchases. Relative indices are a simple way to maintain time-series integrity within each sector index while allowing for the proper weighting of the sectors when calculating the aggregate economic impact ('Weighted Composite Index') of all consumer activities.
4) Relative indexes allow us to expand our sample sizes and market coverage from time to time without distorting the indices themselves. Changes in either market penetration or coverage will significantly change the raw quantities of consumer interest sampled. Relative indices provide a natural and transparent way to internally adjust for changes in our sampling methodologies while calculating each base-100 year-over-year index.
5) Relative indexes of Consumer interest supply a crude day-by-day measure of year-over-year growth or contraction in the Consumer driven portion of the U.S. Economy. If the averaged value of the 'Weighted Composite Index' for a given quarter was 105, a crude 'first order' guess at the annualized growth of the Consumer Sector of the U.S. Economy would be 5%. Similarly, if the averaged value of the same index were 95, someone might guess that the Consumer Sector contracted over that quarter at an annualized rate of 5%.
Note: An index with a prolonged value substantially above or below 100 is experiencing a significant period of compounded (i.e., exponential) growth or collapse. A sustained value of 120 would indicate a compounded 20% year-over-year growth in consumer interest. Similarly, a sustained index value of 80 indicates a compounded 20% year-over-year 'death-spiral' of consumer interest. Since neither of these conditions should be expected to be maintained during normal economic times, any index should have a natural tendency to return over time to 'normal' readings closer to 100.
6) Year-Over-Year relative indexes are inherently seasonally adjusted, without having to resort to any separate and arbitrary 'seasonal adjustment' factors. Our indexes always reflect activity relative to a year earlier (actually, 364 days earlier to compare against the same day of the week). Thus relative year-over-year consumer demand for jet skis and snow boards are properly reflected, regardless of the current calendar reading.
7) Relative indexes allow for the correction of a bias in the raw numbers caused by the gradual overall shift of commerce from brick and mortar stores to the internet. This means that because of the intrinsic growth in internet activities, we must also 'normalize' our indexes to be unbiased by the underlying background growth of internet commerce. To accomplish this we also monitor the overall growth in all forms of internet activity at popular search portals to provide a day-by-day background growth factor that effectively removes the commerce shift bias from our indexes.
Since activities in each of the separate sectors have different impacts on the overall economy, the 'Weighted Composite Index' is calculated in a way that reflects the relative importance of each sector's contribution to the total consumer economy.
This weighting is actually done at a sub-component level within each sector (e.g., separately tallying the 'transportation' and 'hotel' components of the Travel Sector) and weighting those sub-components according to their portion of the total U.S. economy as reflected in the current United States Department of Commerce's National Income and Product Accounts ('NIPA') Tables. Since the relative sizes of the sub-components change daily, the net weightings provided by the NIPA matrix are actually dynamic and change over time.
For example, as mentioned, our Travel sector consists of 'Inter-city Personal Transportation', 'Hospitality' and 'Rental Vehicles' components which are in very different places in the 'NIPA' tables, which in turn are weighted very differently as portions of national economic activity. As the mix between those components changes dynamically day-to-day the net weighting of the Travel sector in our 'Weighted Composite Index' will change accordingly.
As an additional example, each retailer that we monitor contributes to our Retail Index according to the number of transactions observed, even though the types of products that they carry have very different weights in the NIPA weighting tables used to calculate our 'Weighted Composite Index'. For this reason Amazon, Home Depot and Walmart impact the Retail Index very differently than they impact the 'Weighted Composite Index'.
For this reason, the 'Weighted Composite Index' may behave significantly differently from any or all of the individual sectors we separately track. In fact, on occasion the 'Weighted Composite Index' has been either above or below any of the individual sector indexes. If the transaction volume weighted sector indexes diverge from the dollar weighted 'Weighted Composite Index', something interesting is happening in the economy. If the sectors are stronger than the 'Weighted Composite Index', consumers may be very actively spending on smaller "feel-good" purchases, while holding back on bigger items that require a longer term financial commitment. Conversely, if the 'Weighted Composite Index' strengthens while the sectors lag, consumers may have cut back on the smaller expenditures in order to fund the big ticket items. In either case consumer "activity" is diverging from the economic impact of that activity.
Since the 'Weighted Composite Index' better reflects the economic impact and transaction size of Consumer activities, it best represents (and is a truer indicator of) the strength of the overall consumer discretionary economy relative to a year earlier and more closely mirrors how the 'demand' side activity will subsequently flow downstream to the 'production' side of the economy.
Most people are aware that their behavior in the internet is captured and used to provide targeted ads and suggest products that may interest them. We mine similar tracking databases to monitor anonymous macroeconomic tendencies within each of our defined sectors on a daily basis. Our analytical methodologies have been developed and refined over time, and they remain the proprietary core of our business. Unfortunately, we must keep the precise details of our sampling process confidential to protect both the integrity of our methodologies and the security of our data.
Authentication and validation of our data collection process is our highest operational priority, and you may notice occasional multi-day delays in the publication of our indexes as we verify and/or aggregate samples for more statistically significant meta-analysis -- especially in cases where the sample sizes are less robust. From time to time one or more of our data sources may experience a significant service interruption. Although for their purposes (providing targeted ads) the down time may not be critical, we still need to have every day's data captured and accounted for. As a consequence on such occasions we delay publication of the indexes until all of the data is complete and correct, with our end result still being day-by-day indexes -- updated during those incidents several times per week instead of every day.
It is important to note both the immediacy of our results and their scope. We sample consumer activities across the entire U.S. economy, sampling activities in all 50 states. Our data is collected daily, and is generally available in the form of updated indices within several days of the sampling period.
The Sub-Indexes that we highlight graphically on each Sector page generally represent a minor and more specific sub-sample of the activities that we follow for the full sector, and the Sub-Indexes are sometimes specifically chosen because they are different, i.e., they show activities in a sub-section of the full sector that are moving in ways quite different from the full sector.
Sub-Indexes generally reflect a greater specificity in consumer decisions (e.g., a decision to purchase a certain make and model of European Automobile) than the broader Sector (e.g., a desire to purchase a new automobile where no specific brand or model has been selected yet).
Additionally we publish the Sub-Indexes on a weekly basis with a weekly resolution, whereas the full Sector Indexes are tracked daily. The primary reason for this difference in update frequency relates to sample size and the need to increase sampling period to obtain statistically significant data for the much more specific activities involved in the Sub-Indexes. Thus the Sub-Indexes will not have either the resolution or timeliness of the full Sector Indexes.
The full economic impact of any prolonged change in Consumer interest and activities can only be understood by analyzing both the level of our indexes and the duration of any extended deviation from the norm. A one day downward blip in the level of an index may have essentially no effect on an entire economy; but that same level, if extended over a quarter or a year, could be devastating. The index's deviation from 100 is a measure of the current level of growth or contraction in the Consumer Sector of the U.S. Economy, but the consequence of such a deviation can only be understood by summing the daily deviations over the duration of the trend.
What this means is that if you were viewing a graph of our 'Weighted Composite Index', the full economic impact of any prolonged deviation of the index above or below a value of 100 is best measured by the area between the graph line and the horizontal line representing value 100. This area 'under' the curve can be either positive or negative, depending on whether the area is above or below the line at value 100.
To provide a context for the scope or scale index excursions, we calculate the net total area 'under' the curve for constant trailing periods. These periods correspond in duration to a quarter, six months and a year - but they are sliding periods, shifting each day so that they always represent the most recent 91 days, 183 days and 365 days. Thus at any given time we have snapshots of the net positive or negative total impact of trends in Consumer interest over sliding (but fixed duration) time periods.
The 'Percentiles' we provide from these calculations simply rank the trailing 'quarter' or 'six months' or 'year' among all similar duration calendar periods tracked by the BEA since the spring of 1947. A ranking of the 50th percentile means that a given trailing period was absolutely average. A ranking of the 90th percentile means that a trailing 91-Day period would be among the best one-tenth of all quarters since 1947 (if it were a true calendar quarter). Similarly, a ranking in the 5th percentile would place a trailing period among the worst 5 percent of all similar duration calendar periods recorded by the BEA since 1947.
The trailing 91-Day, 183-Day and 365-Day Growth Indexes (or sometimes shown as Growth %) are simply 91, 183 and 365 day (respectively) moving averages for the year-over-year net growth/contraction of the 'Weighted Composite Index'. These are essentially the 'areas under the curve' that are ultimately used to calculate the corresponding 'trailing percentiles' through the statistical comparison of these values with one, two and four quarter growth histories respectively in the U.S. Department of Commerce's Bureau of Economic AnalysisBEA Growth Tables. These 'Growth Indexes' are the year-over-year actual growth from which the 'Trailing Percentiles' are drawn.
The 91-Day Growth Index is closely followed as 'demand side' proxies for major economic statistics. Similarly the 183-Day Growth Index is our demand side measure of whether or not the Consumer stimuli to the economy has been expanding or contracting for the equivalent of two consecutive quarters. Traditional economists use the 'two consecutive quarters' of growth or contraction as a signal for the economy dropping into or recovering from a recession. Our 183-Day trailing 'two quarters' does exactly the same for Consumer activities on the demand side of the economy. If the number is positive, the average reading of our 'Weighted Composite Index' over the preceding sliding 'two quarters' is signalling growth and no 'recession'. If, on the other hand, the 183-Day Growth Index is negative, then the trailing 'two quarters' were in net contraction, signalling a demand side 'recession'.
First of all, it is important to understand what the GDP reports are actually telling us:
Most people hear the GDP numbers and think that the BEA simply taps into all of the cash registers around the country and tallies them up to calculate the quarterly commerce. That would be nice, and it would certainly be far more accurate than what really happens. The BEA methodologies are actually just an extension of a model they started in 1937, with e-mailed questionnaires replacing snail-mailed forms and spreadsheets replacing ledger books. In 1937 the best that they could do was to sample activity at a few hundred factories and try to extrapolate from those statistically small samples to the entire U.S. economy. Times have changed and sample sizes have increased to tens of thousands instead of hundreds, but the basic concept remains: send questionnaires to a minor portion of all U.S. business enterprises and extrapolate the whole economy from those results.
Instead of tallying up commerce, the BEA runs the "mother of all spreadsheets" -- containing complex formulas that try to model the interactions within the whole economy. Picture in your mind the most elaborate, convoluted and opaque spreadsheet known to man. Increase that by an order of magnitude and have it maintained by hundreds of clerks -- none of whom have purview to the entire thing. Now assume that some of the inputs to this spreadsheet are dollar values of transactions, while other inputs are actually quantities of goods counted as they get transported or warehoused -- i.e., no prices/valuations are simultaneously collected, and somehow those goods have to get converted into dollars within the model.
This spreadsheet has thousands of line-item rows representing different parts of the economy (e.g., groceries, automobiles, housing, exported tools, imported oil, finished goods inventories and municipal governments). And it now has hundreds of columns, representing each of the quarters since 1947 (and the years prior to that). At the end of January, April, July and October a new column is added to the right hand edge of the spreadsheet containing the numbers for the most recently ended calendar quarter. The numbers are then "rolled up" for publication in spreadsheets with about 60 rows and published in their quarterly report.
Additionally, as economists, the BEA is obligated to actually have multiple versions of this humongous spreadsheet: one with raw numbers, one with the raw numbers "seasonally adjusted," and one with the numbers corrected for inflation (the so-called "real" numbers). Seasonal adjustments are meant to smooth out the varying levels of commerce for goods that don't have constant year-long consumer demand (e.g., mistletoe, bikinis or home heating oil), but they are notoriously difficult to calculate and maintain (i.e., fudge). They also introduce spurious results when seasonal patterns get skewed (e.g., by non-seasonal weather patterns, one-time events like 9/11 or the growth of a medicinal mistletoe industry).
Inflation corrections are handled by a series of "deflaters" that convert the "nominal seasonally adjusted" current dollars into "real" or "chained" 2005 equivalent dollars. Since different types of goods are inflating at different rates, different "deflaters" are used for different line-item rows in the enormous spreadsheet. Clerks calculating the appropriate "deflaters" aren't necessarily even talking to the clerks counting the inventoried goods, nor for that matter is it even their job to be concerned about what their "deflaters" may be doing to the overall spreadsheet.
What all of this means is that the BEA isn't actually publishing measurements of what is happening in the economy -- they are publishing what is happening in their humongous, convoluted and opaque spreadsheet/model as a consequence of keying in the results from their questionnaires.
The bottom line to all of this is that the BEA is only reporting the results of a huge spreadsheet model of the U.S. economy that collects inputs using techniques first developed in 1937 to target things that mattered to FDR's factory employed constituents. That model samples a small part of the economy and makes enormous extrapolations, seasonal adjustments and price normalizations -- all of which are subject to significant errors. No wonder that normal people on "Main Street" don't really trust government data anymore, because it often just doesn't seem to agree with what they are seeing with their very own eyes.
As a consequence of the above, there are a host of reasons why our data may differ from those published by the BEA. Some of these might be summarized as follows:
1) In order to maintain consistent data series, the BEA is measuring the same types of supply-side data that they first developed in the middle of the previous century in a effort to analyze economic activities that were important at that time. We, on the other hand, are using twenty-first century technologies to measure real-time demand-side consumer activities typical of today.
2) The pace of the mid-twentieth century economy was such that a quarterly update was considered adequate for largely academic pursuits. Our indexes, however, have a daily time resolution and are updated daily (within days of data acquisition) for the benefit of the investing public.
3) The BEA extensively revises their numbers over several months (and ultimately over several years) in order to get the numbers finally 'right'. This may have been acceptable 60 years ago, particularly when used for leisurely academic analysis. In contrast, our daily numbers are final when published.
4) There is a natural lag between changes in demand and when the impact of those changes filter down through the supply chain to factories. If Consumer demand is the fundamental stimulus of most economic activity in the U.S., we are much closer to (or further 'upstream' towards) the source of any changes in the economy.
5) We are measuring only on-line consumer demand, whereas the BEA is tying to measure the entire economy -- including governmental spending, commercial investments and exports. Consumer demand has historically represented 70% of the U.S. economy, but recently unprecedented governmental stimuli and export swings have dominated the changes in the economy, marginalizing the impact of shifts in consumer demand.
6) On-line consumer demand for discretionary durable goods include demographic and cultural biases that may have been intensified by economic changes that have been unevenly shared across the U.S. population. The "X" and "Y" generations that are more comfortable making on-line purchases may have recently experienced a disproportionate share of the recession's economic pain because of lower job security and more precarious credit ratings.
Nevertheless, we feel that investors deserve information that is upstream economically, has daily resolution, isn't noisy or frequently revised, and is measuring what Consumers are actually doing in the current century.
To understand the causes of this divergence we need to look a little more closely at what the BEA tries to measure. The classic definition of the GDP can be summarized in the following equation:
GDP = private consumption + gross investment + government spending + (exports − imports)
or, as it is commonly expressed in algebraic shorthand:
GDP = C + I + G + (X-M)
where "C" (consumer demand) represents about 70% of the entire U.S. economy. With that equation in mind, we offer the following observations:
1) It is important to remember that at the Consumer Metrics Institute we measure only a portion of the "C" in the above equation. In fact, we have intentionally chosen to track a particularly volatile subset of "C" in order to gain signal strength and lead time. Regular visitors to our site know our standard disclaimer: "we capture the discretionary durable goods transactions of internet shopping consumers." We don't track the core non-discretionary items that represent perhaps 90% of take-home pay: groceries, gasoline, utilities, non-discretionary medical expenses and current housing. And in a deleveraging world, you can add personal savings and/or debt retirement to that list -- thus even further marginalizing our data. But -- and this is a really big "but" -- what we measure is by far the largest driving force behind economic growth and new jobs. Furthermore, the demographics of our shoppers makes them (as empirically observed) early trend setters for consumer durable goods, further enhancing our lead times. All of the above tells us that we should lead the BEA's "C" while having an amplified signal that may or may not offset the impact of "I", "G", "X" and "M" when the final calculation of GDP is performed.
(Note that this amplification means that if we were to record a 10% drop in real per-capita consumer discretionary spending on durable goods (excluding residential construction), that drop might result in only a 4% decrease in total consumer goods spending, a 1.4% drop in overall consumer expenditures (the BEA's Personal Consumption Expenditures "PCE" number) and a 1% contraction in the headline GDP.)
2) We think that the BEA's methodologies for imputing "C" are seriously flawed. Their 1937 based focus on factories places their data far downstream from where the real economic action is -- probably 4 or 5 months. We understand why a factory focus was chosen in 1937 (given FDR's constituency and 1937 jobs demographics), but the economy is much more than just factories in 2010. Additionally, the BEA uses a questionnaire approach, which leads to survivor and large firm biases -- not to mention lags and revisions when the data does finally come in. And finally their numbers are "annualized" growth that is then seasonally adjusted; while our our numbers are strictly year-over-year growth, which require no seasonal adjustments.
3) Another problem with using factory data to impute "C" is that the BEA feels compelled to somehow reconcile the downstream data source to upstream demand by tracking inventories as they slosh up and down. Doubly unfortunate is the fact that the BEA's inventory data is very, very late arriving -- and it is by far the largest source of post-2nd revision adjustments to the GDP. So the GDP gets bounced all over the place as inventory building/depleting cycles take place, and our measurements will diverge from the GDP by at least the amount of those swings.
4) In an economy where household leveraging or deleveraging has become commonplace, the relative impact of "C" on the GDP should naturally change. In times of deleveraging, John Maynard Keynes believed that we should simply print new money and crank up "G" to offset the drop in "C". If "G" soars there will have to be some decoupling of the final GDP value from "C."
5) Additionally, "(X-M)" could significantly boost GDP when the value of the dollar is falling, thus causing net exports to grow. Unfortunately, the biggest portions of "M" are either valued in dollars or in currencies defacto pegged to the dollar, and falling values of the dollar don't actually help that part of the equation as much as one might suspect. And since most other central banks want their export goods favorably priced relative to the dollar, there are compelling international reasons to keep the dollar strong -- not the least of which is the dollar's role as defacto world reserve currency and safe harbor during times of global economic distress, which causes international investors and central banks alike want to preserve the value of their dollar denominated assets.
6) Furthermore we have chosen to use as a baseline year the same one (2005) that the BEA uses to build its chained-dollar (i.e., inflation adjusted) calculations. As a consequence, our weightings assume that residential housing is still an important part of the economy. The growth numbers contained in the GDP reports are calculated from the annualization of quarter-to-quarter changes in the economy. So, for example, if the construction and marketing of residential housing was near zero last quarter, a historically minor improvement in raw quantities of housing starts or sales may result in a huge quarter-to-quarter percentage gain, which is then greatly amplified in the BEA's annualization process. Our data, however, will always weight the housing market portion of the economy according to the 2005 NIPA tables.
7) Our data is inherently inflation free -- or "real" using the terminology of the BEA reports. At the very lowest level we measure the year-over-year changes in consumer demand for units of like-kind goods. In a sense we are measuring the number of cars, homes or widgets being consumed by on-line shoppers -- and we therefore do not need to apply any "adjustments" to compensate for changes in price levels. The BEA, on the other hand, has to use elaborate (and often noisy) "deflaters" to convert their "nominal" data into inflation adjusted "real" numbers. As a consequence the BEA numbers from time to time can be taken hostage by aberrant (or "seasonal") inflation factors that result in distorted reporting.
8) In order to normalize our data to compensate for growth of internet based commerce -vs- "brick and mortar" commerce, we have developed an advanced set of "same shopper" metrics (utilizing IP addresses) to separate the impact of new on-line shoppers from true increased actual demand within our previously identified shopper base. One of the consequences of that "same shopper" analysis is that our data is also intrinsically "per capita," corresponding more closely to average household spending statistics than aggregate gross economic data that grows with the population -- as is reported in the headline BEA growth figures. From our perspective the "per capita" data that we provide more closely tracks the true state of the consumers in our economy.
9) The BEA's monitoring of factory data misses to a large extent the "financialization" of the U.S. economy over the past two decades. Our focus on the consumer arguably also misses this phenomena, except to the extent that our consumers leveraged themselves. Where the disconnect with "financialization" is most evident, however, is the discrepancy between our measurements of the U.S. consumer economy and the behavior of the financial markets.
10) It is also possible that some of the biases inherent in our on-line sampling methodologies could cause our data to drift relative to the overall consumer economy. Our data comes exclusively from the shopping habits of on-line consumers, who are demographically different from the profile of the "average" U.S. consumer. On-line shoppers tend to be younger, better educated and less culturally diverse than the U.S. population as a whole. But those same shoppers are also more likely to have lower job seniority and and higher leverage ratios than "average" U.S. consumers, making them more susceptible to widespread downturns in employment or real estate valuations. In such cases our base of consumers will be disproportionately hurt by economic contractions, and our data will drift accordingly.
11) And finally, when our demographics are coupled with the "per capita" nature of our data it is probable that our aggregate data more clearly represents the circumstances of the "median" consumer than the "average" consumer often extracted from the BEA's Personal Consumption Expenditures (PCE) data. If the income gap between the poorest and wealthiest deciles of U.S. consumers is widening, then it is likely that the gap between the mean (or arithmetic average) consumer and the median (or mid-point in the distribution) consumer is also increasing -- as the income gains among the wealthiest push the averages up without benefiting all consumers proportionately. Thus our data may not fully show the impact on the PCE of "wealth effects" (e.g., equity market gains) experienced by elite consumers.
Looking again at the equations above, we can understand that the portion of "C" that we measure should decouple from the reported GDP when consumers deleverage, "G" soars, inventories build, a weakening dollar causes exports to grow, or our consumer base is selectively impacted by economic events.
During bulk of the 2007-2009 "Great Recession" (i.e., prior to the vast governmental stimuli typified by the ARRA) we gained some notoriety for having our Daily Growth Index lead the GDP by a relatively consistent 18-20 weeks. Once the majority of economic stimulus shifted from consumers to the government, that relationship materially weakened. Our methodologies capture only on-line consumer demand for discretionary durable goods, which has historically been the most volatile portion of the GDP. But as a consequence of that focus we do not see the impact of those governmental stimuli that do not flow through consumers. If the governmental stimuli packages that flow exclusively through commercial contractors offset significant changes in consumer demand, the GDP might significantly decouple from our measurements.
By analogy to (American) football statistics, we are only measuring the performance of the starting quarterback for the U.S. economic team. It is possible for a football team to win even though the quarterback is below average -- an overwhelming defense and a punishing running game can compensate for a journey-man quarterback -- but the performance of the starting quarterback is by far the best predictor of a football team's final results. Similarly, the U.S. economy might grow without the U.S. consumer's support, but only with net exports and/or unsustainable governmental consumption. At the current time the likelihood of the U.S. becoming a net exporter is very low, and unsustainable governmental consumption is simply that: unsustainable.
It is also helpful to distinguish between 'leading' and 'predicting'; we have deliberately decided to measure on-line discretionary consumer demand data because it is highly leading, while fully realizing that the volatile data provides amplified signals. Fortunately during most of the 2008 recession the BEA's numbers for the full economy eventually closely matched the discretionary consumer demand portion (that we measure). While we know that we are measuring only one portion of the economy -- the quarterback in the above analogy -- we still feel that those measurements reliably lead the economy as a whole. And it may be unreasonable to expect the BEA's 1937 based measurements -- of those portions of the economy that really mattered in 1937 -- to even ultimately get the numbers right for the current U.S. consumer based economy.
As the saying goes: our numbers are what they are. They are pure daily measures of on-line consumer demand for discretionary durable goods. If consumer demand decisions initiate 70% of all U.S. commerce, we would like to measure that demand as far 'upstream' as possible.
No -- they should not be used to "time" the stock market. We are not investment advisors. We present our indices for what they are: simple measurements of consumer interest in making major discretionary purchases.
The value of our indexes to investors may lie in the frequency and immediacy of the data. For these reasons, our indicators are in a sense the most leading of all common leading indicators - not simply because we are measuring activities which are more 'upstream' economically (and therefore intrinsically more 'leading'), but also because our measurements are generally available sooner and with higher time period (daily) resolution.
Additionally, the 'Trailing Percentiles' may assist investors assess the risks represented by the changes we monitor in consumer interest and activities. For example, our 'Trailing 91-Day Percentile' might give a risk averse investor a leading perspective on the possible severity of approaching economic slowdowns.
Several words of caution need to be said to investors interested in using our information as a guide for their investment decisions:
1) We are measuring primarily consumer activities relating to the largely discretionary purchases of durable goods. This is a major component of the U.S. economy, but not the only one. Other actions by commercial or governmental entities are not necessarily reflected in our samples, nor are non-discretionary consumer expenses (e.g., groceries, utilities, non-discretionary medical care and taxes).
2) Most of the widely followed leading indicators are designed to lead the overall economy, and they are revised on a monthly basis. In contrast, our leading indicators are measuring only consumer activities that haven't been recorded yet by the BEA's measurements, and our time resolution is so detailed that a week's upward movement might not ever materialize as movement within longer term or broader economic measurements. Even if our indicators accurately predict current economic changes, the reported levels of factory activities as a result of those changes will necessarily lag our indicators by months or quarters.
3) Corporate earnings (and therefore the equity markets) can rise or fall even when the economy is moving in the opposite direction. This is because corporations have a great deal of control over their "bottom line" even if their revenues or levels of commerce are at times beyond their control. Drastic cost cutting can raise earnings even as corporate sales decline. And major U.S. corporations often have substantial international operations or sales, making their "bottom lines" also subject to the health of non-U.S. economies and even foreign exchange rate windfalls or losses.
4) Investment markets react to things other than the levels of consumer commerce. Natural disasters or political upheavals can cause market movements that are quite independent of then current fundamental economic activities. Public sentiment towards investment markets can also differ markedly from the actions that those same consumers are making in their day-to-day lives. Thus investment markets can be driven by emotions, events and the news media in ways that are substantially different from the level of consumer activities.
Thus our indexes should not be used as the sole (or even primary) source of information for investment decisions. Nevertheless, our many indexes probably should be considered useful components in the overall investment decision process.
In order to provide our data in as timely a manner as possible, our sampling techniques are electronic and rely on the internet for communications. As a result of our reliance on the very same technologies you are now using to view our results, the consumers whose interests we gauge are, by definition, connected to the internet. In general, our use of only internet connected consumers may bias our results in several ways:
1) Our 'consumer' demographics may be tilted towards higher income and education levels than the entire U.S. consumer base. Our 'consumer' demographics will mirror the demographics of internet users, not the populace of the U.S. as a whole. Thus our results will show the same gender and age bias, for example, as internet users have relative to the U.S. population. Similarly, people with limited access to the internet because of geography or poverty will be excluded from our results.
2) At this time our sampling technologies only function in English. We may expand our techniques in the future, but at present we do not measure consumer activities (even in the U.S.) that are conducted in Spanish, for example.
3) As mentioned above, we do not collect data on purchases that involve little or no discretion on the part of the consumer. These would include utility bills, monthly mortgage payments, essential medical expenses, gasoline and ordinary groceries. These kinds of purchases are core economic activities that involve little or no thought (or discretion) on the part of the consumer. We are interested only in the highly variable (and discretionary) parts of the consumer economy.
4) The reference data used in calculating our 'Trailing Percentiles' is from a table of historical quarters of economic growth and contraction data kept by the Bureau of Economic Analysis (BEA) of the U.S. Department of Commerce. The roughly 250 calendar quarters of data in the table start with the 2nd quarter of 1947 and have a mean annualized growth of about 3.3%, with a standard deviation slightly above 4%. The 60 year time period covered by the table is certainly lengthy, but it may not be truly representative of the U.S. Economy of the past 10 or 20 years. Nor does it remotely represent the entire 20th century. The 60-plus years in the table may be a period of extraordinary prosperity that is not representative of the U.S. Economy over longer (or more current) time periods.
The Consumer Metrics Institute was founded on a simple observation: many 'leading' economic indicators are published, but few (if any) are sufficiently 'leading' to be meaningful to investors. In fact, many 'leading' indicators use the prior month's equity market results as a key component of their indexes. Investors may find their most recent month-end account statements more timely.
To remedy this, the Consumer Metrics Institute has developed (and is continuing to develop) techniques for monitoring 'up-stream' economic activities on a daily basis. The daily consumer sampling process commenced in 2004, and several years of data were required to refine the process and statistically analyze how the timing of our indexes related to other 'leading' indicators, including the equity markets. The 2008-2009 recession provided a final validation of the methodologies and confirmed a multi-month lead relative to other commonly referenced indicators. Additionally, the 2008-2009 event was significant enough to verify whether our trailing percentiles adequately reflected the severity of the downturn. By the summer of 2009 we were ready to release the first results of our ongoing research.
The Automotive Index is indicative of consumer interest in new and used automobiles. It includes interest in car dealerships, auto loans, auto leasing, auto insurance, and automotive accessories.
The Entertainment Index demonstrates consumer interest in the purchase of public entertainment and dining. This index includes live concerts, theatrical presentations, public movie exhibitions, ticketed sporting events, full service restaurants, taverns and lounges.
The Financial Index tracks consumer interest in consumer investment opportunities, including mutual funds, stocks, bonds, annuities, insurance, banks and credit unions. Interest in services related to the banking and financial industries (including securities brokerage, consumer loans and credit reporting) can also be found in this index. The Financial Index also responds inversely to concerns that consumers have expressed about potential credit card defaults or foreclosures.
The Health Index informs us about consumer interest in making discretionary health and medical purchases. Included are such items as weight loss programs, health clubs, spas, discretionary medical procedures (e.g., cosmetic surgery, laser eye surgery and orthodontia).
The Household Index indicates consumer interest in durable household goods, including furniture, appliances, home decorating supplies, home remodeling supplies and services, apparel and jewelry.
The Housing Index captures consumer interest in the purchase of new and existing housing, first time mortgages, refinancing of existing mortgages, and rental homes or apartments. Interest in the engagement of ancillary services (e.g. property appraisals, property inspections, title searches, real estate agents and mortgage brokers) is also caught by this index.
The Recreation Index shows the level of consumer interest in sporting goods, recreational vehicles (motorcycles, snowmobiles, campers and trailers), camping and fishing equipment, boats, hunting equipment, bicycles, exercise equipment, swim outfits (including scuba and surfing equipment), crafts and hobby supplies (including high-end photographic equipment).
The Retail Index directly responds to consumer activities when seeking retail sources for durable goods. The specific types of goods to be purchased may be reflected in other categories, but this index most closely and immediately tracks imminent purchasing decisions that translate into retail sales.
The Technology Index reflects consumer interest in all manner of consumer electronics, including computers, wireless equipment, cell phones, video games, televisions, home theaters, audio systems, personal music devices, music CDs, DVDs, cameras, and downloads of music or video content.
The Travel Index tells us about consumer interest in major transportation (e.g., airline tickets, inter-city bus tickets and inter-city rail tickets), destination accommodations (e.g., hotels and resorts), destination dining, rental cars, tours, cruises, amusement parks and casinos.