Can Wind Speed Data Really Predict Pattern Energy Earnings?
Summary
- A model for using wind speed data to predict Pattern Energy power generation is proposed.
- Results were backtested with 3 years of data and show reasonable agreement between projections and actual data.
- Earnings report and recent announcements indicate Pattern Energy may have difficulty growing in the near term.
In my previous article, I proposed that monthly average wind speed data available from the National Oceanic and Atmospheric Administration (NOAA) could be a predictor of earnings for Pattern Energy (NASDAQ:PEGI) due to its position as a pure play on wind energy in North America. Thanks to numerous comments from readers, I was encouraged to backtest my model and enhance it to produce a prediction of power generation. In this article, I would like to present my results and add some commentary on the recent earnings report.
Methodology
Wind speed data was obtained from NOAA monthly average wind speed maps. For each month, two maps are available: a map of the measured average wind speed and a map of the wind speed anomaly. The average wind speed map is exactly what it sounds like: the absolute value of the wind speed in meters/second. The wind speed anomaly map shows how much the average wind speed for that month deviated from long term average wind speeds (over 30 years). Using an image editor, a second layer was added to the map on which was plotted the approximate locations of Pattern Energy wind farms. This layer was then copied onto each map analyzed to maintain consistency. Here is what the maps for January 2018 look like:
Source: NOAA, Pattern Energy December 2017 Investor Presentation
Using the legend provided, numerical values were assigned for the average wind speed and wind speed anomaly at each location. If a location was solidly within one color the midpoint of range for that color was used. If a location was on the border between two colors, the value assigned to the transition was used. For locations in Canada values were extrapolated based on nearby wind speed patterns in the U.S.
Source: Author, data compiled from NOAA wind speed maps
To predict power generation for a quarter, the percentage difference in wind speed is cubed, averaged across the three months, and weighted by the capacity of the site (MW) relative to the portfolio.
A Few Caveats
Before I delve into the results of the backtesting, I’d like to make a few comments on the validity of this model, which includes a lot of assumptions, extrapolations, and sources of error. Beginning with the data source: while the NOAA maps provide a broad overview of the wind conditions over large areas, they do not have the granularity to show the specific conditions at each wind farm. Local wind speeds may vary greatly from the averages for a large area. Second, since the locations of the wind farms were manually plotted, there may be some accuracy error. At this scale, even a millimeter in difference could represent tens of miles of distance which could results in very different wind conditions. Third, the data lacks precision – is a site with an anomaly color of yellow actually experiencing 0.01 m/s above average or 0.49 m/s above average? To somewhat mitigate this I used the midpoint of the range, but that is an arbitrary number which certainly introduces error. This error is further magnified by the calculations to derive projected power. A 0.25 m/s anomaly may not be that significant when the average wind speed is 7 m/s, but is a much bigger deal if the average wind speed is only 3 m/s.
NOAA data is derived from wind speed measurements obtained at weather stations. These stations typically measure wind speed at 10 meters above ground. However, wind turbines are typically at heights of 80 meters or higher in order to take advantage of stronger, steadier winds. Ground level winds can be used to estimate wind speeds at higher altitudes using the wind profile power law, but this introduces another source of error into the model.
Another source of uncertainty is the frequency distribution of wind speeds. Although we have a monthly average, wind speeds change on the order of minutes or even seconds. Because the power production depends on the cube of the wind speed, the frequency distribution is more important than long term averages. For example, a turbine experiencing 50% winds at 6 m/s and 50% winds at 4 m/s would produce more power than a turbine experiencing 100% winds at 5 m/s. Related to this, wind speeds above the turbine rated speed or below the cut-in speed do not provide incremental power generation, even though they affect the wind speed average.
As I have mentioned, the data includes extrapolations for wind speeds in Canada, which may or may not be accurate. Additionally, data is missing for the Santa Isabel, El Arrayan, and Meikle sites; for the purposes of the model, these sites are assumed to be producing at levels consistent with long term averages.
All of this is to say that this is by no means a refined model and should not be relied upon to produce a precise or accurate prediction of power generation. Rather, it can be used to gain insight into trends and provide an estimate as to the availability of wind resources for power generation.
Backtesting Results
The model was backtested starting from Q1 2015, which is the first quarterly report I could find which included commentary on produced power relative to expected long term averages. Note that in previous years, management only provided vague references as to the performance of the portfolio. Also of note, during parts of 2015 and 2016, North America was under El Nino conditions which resulted in wind speeds far below average. Of more interest, beginning in Q2 2017, a section on operating metrics was included in quarterly 8-Ks which provided much more detailed information about the actual performance of the portfolio relative to long term expectations. Without further ado, the results:
Source: Author, data compiled from NOAA wind speed maps, Pattern Energy press releases, conference call transcripts, and 8-K reports
A summary is below:
Source: Author
The model seems to have trouble with periods when the wind is extremely low or high. Results for 2015 seem all over the place, but are slightly better for 2016. For 2017, the model seems reasonably accurate. This may be due to fewer wind farm sites in earlier years combined with El Nino effects allowing missing data or extreme deviations at individual sites to skew the results. Overall, the model seems useful for predicting whether production will come in above, at, or below expectations based on long term averages.
As a side note, Pattern Energy has yet to report production coming in above expectations; this data seems to corroborate that wind speeds were, in fact, below normal the past 3 years (management is not lying to us about underperformance):
Source: Author
Side Note #2: Several sites (Panhandle 1 and 2, Post Rock) seem to be perennial underperformers. Just bad luck? Perhaps these should be considered for “asset recycling.”
Going Forward
A model is only as good as the data you feed it, and the current data source certainly leaves much to be desired. I am looking into sourcing data from individual weather stations nearby the wind farm sites, which would allow for more accurate wind speed data, and hopefully, daily, or even hourly average wind speeds.
Pattern Energy’s recently announced acquisition of assets in Japan certainly throws a wrench in the works with regards to the model – we now have to add solar radiation levels and wind speeds in Japan to the list of data not readily obtainable. Fortunately, they are still a relatively small portion of the overall portfolio, so I think, for the time being, North American wind speeds will remain the largest factor in predicting power production and earnings for Pattern Energy.
About That Earnings Report
My assorted thoughts and reactions to the earnings report and conference call:
- 2017 dividend payout ratio was approximately 100% of CAFD – slightly concerning, since management has been trying to reduce this ratio. Given the poor wind speeds, this is understandable. Addition of the Japanese assets and start of commercial operations at Mont Sainte-Marguerite should alleviate this a bit. Not raising the dividend is a good decision, but assuming it is held at this level, Pattern Energy needs to hit midpoint of guidance just to maintain 100% payout ratio (98M shares X $1.688 annualized = $165M, guidance 151M-181M)
- Liquidity is getting low… – Pattern Energy entered 2018 with $666M total liquidity and has since committed 132M to acquire existing Japanese assets plus $194M to construct the Tsugaru project (80M due once financing closes, remainder due 2020). In addition, Pattern Energy expects to continue to fund capital calls for its stake in Pattern Development (300M expected total commitment, already funded 60M in July, 7M in Dec, 35M in Feb)
- …but can’t raise capital – at current stock prices, management has pretty much ruled out issuing common equity. Additional debt is not really an option either, if they want to maintain a 3X-4X leverage ratio. A dividend cut would be the easiest source of cash but would absolutely ruin Pattern Energy’s reputation and certainly lock it out of issuing equity at any reasonable price for the foreseeable future – definitely not advisable. As such, management is exploring more creative options for increasing liquidity such as selling assets, issuing preferred shares, etc. Given the current liquidity situation, I expect at most 1 more (small) dropdown this year.
- Common stock now yields approximately 10% - this is better than long term average returns for equity. Pattern Energy’s revenue stream is basically guaranteed at least for the next decade (weather permitting). If management did absolutely nothing from here on out (no growth, distribute all CAFD) I would be happy with those returns. That’s not to say the stock can’t get any cheaper, but this is certainly an attractive entry point. I plan to initiate a small speculative position.
This article was written by
I am the curator of the Dividend Champions list, a monthly publication of companies with a history of consistently increasing their dividends. My primary investing focus is in deep value and dividend paying stocks, but I am constantly exploring alternative strategies. I have a Ph.D in Chemistry from Rice University and have earned the CFA Institute Investment Foundations certificate. I am a contributor to The Dividend Kings marketplace service.
Analyst’s Disclosure: I/we have no positions in any stocks mentioned, but may initiate a long position in PEGI over the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.
Seeking Alpha's Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.