Using a financial markets’ trend-analyses tool to assess temporal trend-changes in global surface temperature anomalies (GSTA).
Guest essay by David Dohbro
Heated debates (pun intended) are currently on going regarding if the Earth’s surface temperatures continue to rise, have remained steady, or are decreasing over the past decade or so. To argue for or against any of these three possibilities, pundits often use (linear) regression lines drawn through parts of the different temperature anomaly data-sets that are publically and freely available (GISS, HadCRUT, NCDC, RSS, UAH) to proof or disproof any or all of these possibilities. The problem is that global surface temperatures are none-linear, stochastic in fact; meaning they are dependent on many (random) variables and cycles each operating on many different spatial and temporal scales; natural and possibly man-made alike. Examples are solar activity, volcanic activity, oceanic cycles such as ENSO, PDO, AMO; night/day cycle, seasonal cycle, trace-gasses, cloudiness, etc. Given the nature of the data, the best representation of a temperature trend over time is therefore by using a stochastic time-series trend analyses of the entire data set.
One of the industries where non-linear trend analyses are and have been done over many years is the financial industry. Reason is that asset prices, for example stock and bond prices, are dependent on many variables; are stochastic, and follow non-linear cyclical patterns. In addition, financial markets may often exhibit a directionless trend in time (See Fig. 1; blue horizontal line). However, within such type of larger scale trends smaller scale trends (prices increase and decrease) occur, and financial decisions to either buy, sell or hold assets based on these trends of different time scales need to be made to ensure maximum profits and minimal losses. A rather important task considering we are talking about a daily multi-trillion dollar industry where having accurate and reliable decision tools are obviously paramount.
The Moving Average Convergence-Divergence (MACD) indicator was therefore developed as an additional tool for investors to provide easy-to-interpret (buy and sell) signals, as well as the direction of the price-trend over time. It is a trend-following signal indicator based on three exponential moving averages (EMAs). The MACD indicator consists of a “MACD Line” and a “Signal Line” (See figure 1; the black and red line, respectively). In this case, the MACD Line is calculated by subtracting the 26-day EMA from the 12-day EMA (See figure 1; the blue and green line, respectively). The Signal Line is the 9-day EMA of the MACD Line. Plotting the MACD Line and Signal Line together with the price data shows how the crossing of these two lines identifies “buy-“ and “sell” signals (See figure 1; the corresponding vertical arrows when the two lines cross), while the direction of the MACD Line identifies the corresponding price-trend. Because the MACD simply subtracts a longer EMA from a shorter EMA it is independent of the nature of the data-set and can be applied to any stochastic (time-series) data set for identification of signals and trends. Theoretically the MACD can thus be applied to global surface temperature anomaly (GSTA) data as well.
Here the MACD is applied to HadCRUT4 data because it is the longest continues data set on record available. First the 12 and 26-year EMAs were calculated from this data, and then subtracted to obtain the MACD. The 9-year EMA was then calculated from the MACD. Both lines were then plotted in the same graph, and the graph placed below the temperature data-set graph on the same time-scale as is done in financial charts (Figure 2). It follows that the MACD of the temperature data peaked or bottomed and then reversed in several instances –see blue vertical lines (Figure 2)- indicating a change of trend in global temperature anomalies; either GSTAs started to increase (~1911, ~1976) or decrease (~1879, 1945, and the latest 2007).
The actual “buy” and “sell” signals (orange arrows) occur a year or two later, because the MACD is a lagging indicator (it is based on longer time-frame moving averages). Note that each and every time these peaks, bottoms and signals occurred in the MACD indicator, temperatures did peak or bottom and subsequently a trend-change occurred: e.g. an increase in GSTA became a decrease and vice versa; no exception. In addition, the MACD also clearly and undeniably identifies the uptrend in temperatures from the mid 1970s till to early 2000s; thought to be the result of mankind’s CO2 emissions; aka anthropogenic global warming (AGW). These “pivot points” validate the yearly-MACD (12, 26, 9) in that it can correctly identify changes in the trends of global surface temperature anomalies reported by HadCRUT4. More about this in detail later.
Now that the MACD-method has been validated we can take a look at the latest signal, which occurred in 2007. The MACD peaked then and has been steadily declining. In addition, the Signal line crossed the MACD in 2008; a “sell” signal occurred. Moreover, the MACD and Signal line are now both pointing down since several years indicating that the temperature trend has changed and the new trend is now down (decrease). Other items of interest that can be deducted from the MACD analyses are the following (See Figure 3):
1) The time-periods between peaks and bottoms in the MACD – blue vertical lines –are of almost identical length (red solid horizontal arrows are of identical length)
2) The increase in MACD (green dotted arrow) is about the same for both periods with increasing GSTA (1911-1945; 1976-2007)
3) The decrease in MACD (yellow dotted arrow) is about the same for both periods with decreasing GSTA (1879-1911; 1945-1976)
What can we learn from these 3 observations? Apparently there are 4 cycles in the current HadCRUT4 data, which suggest GSTAs are now in the next ~32yr cooling period (like any model, we have to work with the data we have and use the past to predict the future). Namely, the MACD of the HadCRUT4 data set finds the following dates with corresponding max and min GSTA values
· max 1879.2 (-0.094), min 1911.7 (-0.362): 32.5yr period
· min 1911.7 (-0.362), max 1945.7 (+0.186): 34.2yr period
· max 1945.7 (+0.186), min 1976.7 (-0.310): 31.0yr period
· min 1976.7 (-0.310), max 2007.0 (+0.829): 30.3yr period
The dates with the actual max and min GSTA values are:
· max 1878.1 (+0.403), min 1911.1 (-0.774): 33.0yr period
· min 1911.1 (-0.774), max 1945.6 (+0.362): 34.5yr period
· max 1945.6 (+0.362), min 1976.2 (-0.439): 30.6yr period
· min 1976.2 (-0.439), max 2007.0 (+0.829): 30.6yr period
The ~32 yr period/cycle; which is an average of these 4 trends becomes apparent, and the MACD does a very good job in determining the dates with the max and min GSTA values. Having determined these dates one can then apply -if one would like to do so- linear regression for each period to determine a slope. Using the actual dates of max, min GSTA values the slopes for each corresponding period/cycle can be determined
· 1879 to 1911: -0.0076°C/yr, R2=0.18 (stat. sign.)
· 1911 to 1945: +0.0141°C/yr, R2=0.52 (stat. sign.)
· 1945 to 1976: -0.0020°C/yr, R2=0.02 (stat. not sign.)
· 1976 to 2007: +0.0193°C/yr, R2=0.64 (stat. sign.)
Using the MACD-determined dates of max, min GSTA-values the slopes for each corresponding period/cycle can be determined
· 1878 to 1911: -0.0066°C/yr, R2=0.15 (stat. sign.)
· 1911 to 1945: +0.0136°C/yr, R2=0.50 (stat. sign.);
· 1945 to 1976: -0.0022°C/yr, R2=0.02 (stat. not sign.)
· 1976 to 2007: +0.0186°C/yr, R2=0.62 (stat. sign.);
It follows, the MACD-determined slopes for each cycle are in very good agreement with those based on using the actual max-, min-GSTA values and dates, showing -again- how accurate and useful the MACD-model is. Point is that stochastic trend and cycle analyses clearly finds periods of about equal length where temperatures rise or decline. The latest cycle, until 2007, indeed saw temperatures rise more rapid, albeit the difference is small, than the previous warming cycle (0.019°C/yr vs 0.014°C/yr; both actual and MACD-determined).
Finally, regression analyses of the data from 2007.0 till 2013.4 shows a slope of -0.002°C/yr and an R2=0.001. Although likely ~25yrs of data for this cooling cycle are still lacking, hence the low R2-value, the slope is already similar to that of the previous cooling cycle. With continuous increasing atmospheric CO2 concentrations since at least 1958 the case can therefore be made that CO2 can not be the main driver in changing GSTA. Instead, the rather similar rates of increases and decreases in GSTAs for the by the MACD identified cycle time-frames, suggest that cycles of around 32 years in length on average, and possibly fractions and multiplications thereof, can explain the observations entirely. The influence of such 30 cycles on Earth’s climate and global temperatures has been reported; e.g. ENSO, AMO, and PDO cycles,,, sea level cycles, length of day / atmospheric circulation index cycles, solar cycle(s), and planetary cycles. Contrary, these ~32 year cycles are not in sync with global human population/economic activity or to global CO2 concentrations. The latter, instead, increases unabated since 1958.
If the current cooling trend is true and applying the ~32yr cycles, it suggests that GSTA should decrease until the late 2030s early 2040s by on average 0.15°C (between 0.06 to 0.24°C) before another warming cycle may commence. Such a cooling trend into the 2030s has been predicted previously.
To conclude, this data-analyses tool suggests objectively and without any adjusting, transformation, fitting, “cherry picking” or other means of data manipulation, that GSTA have likely peaked and are now decreasing; a change of trend has occurred. This technique also over comes IPCC’s claim that “Due to natural variability, trends based on short records are very sensitive to the beginning and end dates and do not in general reflect long-term climate trends.” as the more data the better.
 Developed by Gerald Appel in the late 1970s. The MACD calculates the difference between two trend-following moving averages; this difference is termed a “momentum oscillator.” The longer period moving average is subtracted from the shorter period moving average to calculate this parameter. As a result, the MACD is an indicator of trend. The MACD fluctuates above and below a zero line as the two individual moving averages converge, cross and diverge over time. See also: http://stockcharts.com/school/doku.php?id=chart_school:technical_indicators:moving_average_conve
 Often the 12, 26 and 9-period EMAs are used, where the period can be any suitable time interval from seconds to days to weeks to months and years.
 Giese B.S., Ray S. 2011. El Niño variability in simple ocean data assimilation (SODA), 1871–2008. Jounral of Geophysical Research, 116, C02024, doi:10.1029/2010JC006695.
 Knudsen et al. 2011. Tracking the Atlantic Multidecadal Oscillation through the last 8,000 years. Nature Communications, 2:178 | DOI: 10.1038/ncomms1186)
 Chambers et al. 2012. Is there a 60-year oscillation in global mean sea level? Geophysical Research Letters, 39 (18), DOI: 10.1029/2012GL052885
 UN Food and Agricultural Organization (FAO), 2001. Climate Change and Long-Term Fluctuation of Commercial Catches. ftp://ftp.fao.org/docrep/fao/005/y2787e/y2787e01.pdf
 Scafetta, N.,2010. Empirical evidence for a celestial origin of the climate oscillations and its implications. Journal of Atmospheric and Solar-Terrestrial Physics, doi:10.1016/j.jastp.2010.04.015.
 Landscheidt, T. New Little Ice Age instead of global warming. Energy and Environment 14, 327-350, 2003.