Reader rsm is perpetually concerned about the precision of economic data, most recently here in re: gasoline prices.
Why am I paying 12-15% more than the graph? What possible reason do you have for not including standard errors? Would error bars stop you from making all these claims about noise?
Below is the price of regular gasoline, plus/minus 1.96 standard errors.
Figure 1: Price of regular gasoline, $/gallon (black line), +/- 1.96 standard error band (blue gray lines). NBER defined recession dates shaded gray. Source: DOE EIA (accessed 12/17/2021), NBER, and author’s calculations.ec
In my view, reporting of standard error bands does not change one’s views regarding the course of gasoline prices.
Note that the webpage for these data comes up in a google search first time around when one puts in search words “gasoline” “standard error” “energy information”…
Update 12/20/2021:
http://www.asasrms.org/Proceedings/y2018/files/867118.pdf for comparisons
There is an old adage – do not feed the trolls. I appreciate the point you are making here as rsm’s comment on this is like all of this comments. Uninformed wastes of time. But can we all make a point of just ignoring him? Hopefully he will get bored and go nag some other blog.
rsm: I’m just accumulating material for the day rsm‘s identity is revealed, so we know who is exactly how ill-equipped to understand the world.
Menzie, where can I find the rules for your blog comments?
Relax dude – your serial lies in defense of a traitor (Trump) has not gotten you banned – even though your destructive lies should have put you in jail.
Very nearly since starting reading this blog. I have been of the impression (right or wrong) that Menzie is in possession of one of the attributes held by many (most??) of the greatest minds of all history: TOLERANCE.
The Chinese have some old saying about “When you open up the windows for fresh air the flies also come in” (paraphrasing). Do you think since Deng Xioaping opened up Shenzhen, Guangdong, Fujian, and onward they regret it now?? Try to tolerate a few flies and you might find it benefits your brain.
The link provides how the EIA comes up with its data in incredible detail including:
‘Every Monday, EIA collects information on retail prices for regular, midgrade, and premium grades of gasoline from a sample of retail gasoline outlets across the United States using Form EIA-878, Motor Gasoline Price Survey Schedule A. The weekly survey is designed to collect data on the cash price offered at the pump (including taxes) to consumers for each grade of gasoline. The data collected represent the price as of 8:00 a.m. local time on Monday, for the self-serve price except in areas having only full-serve, and the cash price except for outlets accepting only credit cards.’
This is clearly a national average. Regional differences exist in part because these prices include the excise taxes which EVERYONE with brain knows varies widely by state. Everyone except rsm as his complaint was that he was paying more than the national average. Now I do not know where this annoying troll lives but EVERYONE knows people in Cali or NY pay more than people in other states. Everyone except the obsessed idiot rsm.
These are standard errors of actual gasoline prices based upon a sampling survey, not standard errors associated with oil or gasoline futures. It is a meaningless analysis, unless you are questioning whether the EIA knows how to survey gas stations.
Steven Kopits: (1) Yes, I know what a standard error from a survey is, and that is what rsm wanted. (2) I don’t know what a standard error for a futures is necessarily (how you calculate kind of depends on what is the population moments are), but I do know what a RMSE and a MAE are for futures are. So this post is not addressed to you.
By the way, aren’t you the one that didn’t think confidence intervals were always of interest, and furthermore misunderstood what a confidence interval was? Why yes you are – do I have to dig up the post where I laid that all out?
That’s true. I am not always a fan of confidence intervals.
If the historical data is relatively stable compared to future events, then confidence intervals or the like can be useful. So, for example, if we take traffic on the George Washington Bridge and adjust for weather, time of day, day of year, and weekend/holiday, then I think a confidence interval is a useable piece of information. You can make decisions on that basis. (Does Waze do this? They have been screwing me over lately.)
However, if the data is unstable or not well understood, if the methodology is new or not well understood, and if the exceptions to the general rules are not well understood, then confidence intervals can provide a false sense of confidence. For example, if I fail to adjust GW Bridge numbers for time of day, then confidence intervals will be effectively useless. So if I took average transit times on the bridge, but crossed at 3 am, then I would likely be too pessimistic about transit times.
I find confidence intervals are often used for political purposes, to project certainty where methodologically, that overstates the case. Sometimes materially. Hurricane Maria in PR serves as a case study.
Steven Kopits: Well, we have developed tools to deal with time varying volatility. Think ARCH…
Uh oh, Menzie. Guess Buz Brock really is no longer showing up on campus (and in fact I know it has been some time since he has, although I know people actively pushing for him to get the Nobel Prize). But, and I am not going to get into the alternatives here as this really does get into esoteric stuff above what most readers here know about, but, seriously, ARCH is maybe not the best way to deal with all of this, although it is better than nothing.
BTW, I apologize to you for getting so mean to the vast majority of readers here, who I am sure are mostly decent people in their actual lives. But I also suspect you are not all that unhappy that I am slapping the faces of all these moral cowards. Heck, I and my wife have both been arrested and tortured by authoritarian foreign governments, but we operate publicly under out own names. Just what are these worthless moral cowards covering up?
Of course, you can adjust time series data for all kinds of conditions. But was it done, and done properly? And sometimes, you really don’t have the option given real world data. The sample sizes are too small, or not entirely clear in their meaning, or perhaps collected at different times. As a result, there are myriad issues relating to the real world conditions in which data are collected, analyzed and interpreted.
Statistics are helpful if the data is stable over time. Sometimes, many times, it’s not. At certain times, it will be apples to oranges. Oil prices are a case in point. The EIA has these stupendously wide error bands around their central estimate. Their 95% confidence interval for 2023 WTI is $35 to $160 / barrel. What is that useful for? What information does it convey at all? What kind of action would that inform? If it has no use, then it’s just a pointless card trick, a demonstration of technique without substance. Why show it at all?
Similarly, the mean estimate of mortality in the Harvard PR study was off by a factor of three! I could have come out with a much better estimate doing back of the envelope — which I did! If you are using a methodology which is new, poorly understood and unproven, you have to be really, really careful. And the Harvard team was not. They were, in fact, extraordinarily sloppy, and yet that number reverberated in the media for weeks. If your methodology or data is garbage, then your CI’s are also garbage. Presenting statistical analysis is more than the mindless running of a statistics package and publishing the results.
So, when the data is relatively stable over time, large in number, well understood, has consistent meaning, has consistent drivers with a well established collection and processing methodology and is properly sub-segmented for things like season, time of day, location, weather, etc, then I think confidence intervals are useful. For example, if we were talking about US oil consumption, which is pretty stable, then CIs are appropriate (and indeed, the EIA has done a pretty good job of forecasting both US and global oil consumption). On the other hand, if we are talking about WTI futures — which are driven by a large variety of hard to predict, and even measure, variables — then CIs are useless at best and misleading at worst. In my opinion, CIs should be avoided as a general rule in those cases.
https://www.eia.gov/outlooks/steo/
EIA will covering SPOT markets for gasoline not future markets for oil. So WTF is your little point here?
was covering. Sorry. Now had Stevie taken the time to READ that EIA link, it was clearly surveying gasoline stations. Come on folks – if you are going to criticize a post, take the time to actually understand what it said. GEESH!
https://fred.stlouisfed.org/graph/?g=If40
January 15, 2018
Prices of US Diesel and Regular Gas, 2017-2021
(Indexed to 2017)
https://fred.stlouisfed.org/graph/?g=JZ24
January 15, 2018
Real Prices of US Diesel and Regular Gas, 2017-2021
(Indexed to 2017)
When Chinese regulators responded:
http://www.news.cn/english/2021-12/03/c_1310349895.htm
December 3, 2021
China to cut gasoline, diesel retail prices
BEIJING — China will cut the retail prices of gasoline and diesel starting Saturday, the country’s top economic planner said Friday.
The prices of gasoline and diesel will go down by 430 yuan (about 67.46 U.S. dollars) per tonne and 415 yuan per tonne, respectively, according to the National Development and Reform Commission….
Dollars per tonne? Good grief. Oil is prices in terms of barrels with each barrel being equivalent to 40 gallons of gasoline. We appreciate the conversion of yuans to dollars but it would be nice to tell readers how a tonne translates into barrels.
http://www.news.cn/english/2021-12/17/c_1310379660.htm
December 17, 2021
China’s installed capacity of renewable energy expands in Jan-Nov
BEIJING — China’s installed clean energy capacity reported rapid growth in the January-November period, data from the National Energy Administration (NEA) shows.
By the end of November, the country’s installed wind power capacity had soared 29 percent year on year to 300 million kilowatts, and its solar power capacity had reached 290 million kilowatts, up 24.1 percent from a year ago.
China’s total installed power generation capacity stood at 2.32 billion kilowatts, growing 9 percent year on year, the NEA data shows.
A circular jointly released by the Communist Party of China Central Committee and the State Council said that the country’s proportion of non-fossil energy consumption should increase to around 20 percent during the 14th Five-Year Plan period (2021-2025).
http://www.news.cn/english/2021-12/10/c_1310364612.htm
December 10, 2021
In the January-November period, NEV sales amounted to about 2.99 million units, surging by 166.8 percent year on year and accounting for 12.7 percent of the total sales of new vehicles., the CAAM data showed.
China aims to raise the proportion of NEVs in its sales of new vehicles to 20 percent by 2025, according to a development plan for China’s NEV industry released in 2020.
http://www.news.cn/english/2021-12/10/c_1310364017.htm
December 10, 2021
Tesla’s Shanghai plant delivers over 400,000 vehicles in first 11 months of 2021
[ Chinese vehicle makers have gained importantly in NEV (new energy vehicle) production and sales, while GM and Toyota have lagged behind. ]
This site tracks gasoline prices by station daily. Not sure how they get to their national average number, but I think it reflects current prices more accurately than Figure 1. https://www.gasbuddy.com/charts
Just talked with an acquaintance of mine in northern Georgia where gas prices are typically lower than many states and he said gas prices were hovering around $3.15 to $3.20 per gallon of regular. He’s a little off the high volume area where prices are more like $3.05 per gallon. Prices in lower Michigan are about $3.30 per gallon although some chains like Kroger and Costco seem to be discounting a bit for their shoppers.
https://www.gasbuddy.com/gaspricemap?lat=38.822395&lng=-96.591588&z=4
$2.90 seems to be the lowest price available; certainly not a national average, much less $2.80.
Bruce Hall: 5 seconds on Google Scholar reveals information on differences in collection/construction of the indices from GasBuddy vs EIA. See https://pdfs.semanticscholar.org/edef/3c1c3ee12053d6fbf3bacc9f068e6b5c2612.pdf .
One is sampled, one is crowdsourced.
Nice comparison. I really like these two bullet points re Gas Buddy:
“Members earn points updating a gas prices; redeemed for prizes in raffles.”
Hey Bruce – start reporting over and over and maybe you can win one of those prizes!
“Automated algorithms” used to detect “obviously wrong information.” Users can also report incorrect
information
Gas buddy is a lot like Wikipedia. Same old Bruce – touting junk as the most reliable source!
pgl, you’ll say anything to have a snarky reply whether or not it is relevant.
Crowdsourcing is what Waze used to differentiate itself from Google Maps, Apple Maps, and Mapquest. The information was so much more superior to that Google actually bought Waze to gain access to the data and methodology (my daughter-in-law was advertising manager for Waze at the time). Real time data is superior to daily sampled data for many applications, but you wouldn’t know that locked away in your basement.
Now that Menzie has finally “seen the light” that GasBuddy is the best source for gas data points and that local Oklahoma prices are the natural barometer for near future national gas prices, I can now rest easy that my crown as the trendsetter for econometrics has not been shoved off my head. Alan Krueger and his damnded durned “replicable research”…….. Hmmmmmph!!!!
[ that’s a joke Menzie ]
Just curious if “crowdsourcing” means less reliable than “sampling” in your estimation. I’m guess “crowdsourcing” provides close to a “population” input versus a “sample”. If so, it would seem that “crowdsourcing” would be much more accurate for gauging the real change which seems to be the conclusion of the pdf document for which you provided a link (see pages 21-22).
Bruce Hall: Then you’d be wrong. Crowdsourcing might provide a bigger sample and hence might be more accurate, but it won’t necessarily provide the entire universe of observations (which would be the population). In order to know what was better, you would need to know some of the characteristics of the respondents. (Are the people with gasbuddy apps a representative sample, e.g.?)
Menzie (and of course pal,
If you don’t like GasBuddy (crowd sourced real time data), will you go with AAA? Their data shows $3.30 per gallon national average for regular gasoline… not too different from $3.29 from GasBuddy, but significantly different from the ~$2.80 you are showing.
https://gasprices.aaa.com/state-gas-price-averages/
No? Then how about the EIA?
https://ycharts.com/indicators/reports/us_weekly_retail_gasoline_and_on_highway_diesel_prices
I think you should check your data and question your sources.
… (and of course pgl)… autocorrect thinks “pgl” is an error, LOL!
And it usually is correct: “autocorrect thinks “pgl” is an error, LOL!” 😉
Bruce Hall: I’m not dismissing other sources. I just happen to know how the sample is conducted when undertaken by EIA. EIA also has a consistent approach *over time*, so one can think of the data as being comparable. I assume AAA maybe the same.
I urge you to think about what a variance is (and while we’re at it, how a standard error compares).
Menzie,
I think you might want to consider your data as an “outlier”. That rapid crash of gasoline prices just did not happen. When the model tells you the sun is shining and you are getting soaked, you might not want to claim that the rain is wrong.
Bruce,
What prices are now in this or that locality is of absolutely zero interest. What is of interest is have they been changing recently, and if so, in what direction and by what amount? in short, are all the media stories still blathering about gasoline prices driving this awful inflation that is the worst thing we have ever seen correct or not? Differences of opinion seriously abound, with today Lawrence Summers in WaPo predicting higher rates of inflation, focusing on housing and saying not a word about gasoline prices, while Dean Baker says we might see deflation for the coming year, although more likely just much lower positive inflation than we are seeing now.
I have been calling for gasoline prices to fall since before anybody else here was, and it looks like they are doing so, at least somewhat. So, are those prices you so assiduously reported to us higher, lower, or the same as what they were say last week or a month ago? That would be useful information.
“What prices are now in this or that locality is of absolutely zero interest.”
Then again Uncle Moses keeps reminding us he can find gasoline at far less than $4 a gallon!
In NYC before they raised excise taxes in NJ, I used to hear people bragging about how they refilled before crossing the Hudson River!
Gas buddy? I guess you get your interest rate forecasts on Instagram.
Rig count up 4 oil, down 1 for gas. There is some week to week variation (just based on pad moves and the like). But still seems to fit with a trend of gradual rig increases.
https://rigcount.bakerhughes.com/na-rig-count/ (click on 4th link down, PDF summary)
At 475 oil rigs, I think we are sized for continued moderate growth in oil production, maybe about 0.5 MM bopd/yr. I think about 425 is needed for equilibrium. We have been drawing down DUCs, but should have exhausted much of the non-rotten DUC count. Going forward, I expect to have a pretty standard 6-7 months of working inventory reasonably soon. Especially as frac spreads tend to roll off more in the winter, than rigs (based on weather issues).
Gas seems sized for about status quo, maybe slight growth based on associated gas from oil volume increases. I think of 100 gas-directed as about equilibrium amount, with associated gas stable. But oil and gas are co-products.
Note the Canada drop in rigs is normal seasonal. Expect a couple weeks of even bigger drops coming up. For some reason, Canadian rigs are more likely to take a Xmas break than US ones. See this every year. Will build back in JAN. (There is also a spring melt drop in Canada in APR-MAY.)
Big weekly drop in frac spreads. Compiler thinks this is seasonal. (I donno, we only have a few years of compiled numbers, not like the 3 decades of rig count info. Plus, 18, 19 both had price trends down in DEC (20 the opposite).
https://www.youtube.com/watch?v=iiBvE5zVt68
Can you show your work? Why must the data be in Excel?
Don’t you really want something like the “Regular Gasoline Prices” chart in https://www.eia.gov/petroleum/gasdiesel/ ?
What is the range on that chart, which shows different US regions?
How is the standard errors you graphed so small? Did they trim means to throw out outliers, thus dismissing my lived experience as unobserved?
Have you heard of ergodicity? What does your graphed mean and tight confidence intervals have to do with my actual experience paying gas prices? Why did one region’s prices go down as others went up? Why are some regions choppier than others?
$3.20 to $4.20 is what, plus 30% on the top confidence interval? Shouldn’t the real error bars be much, much wider than the figures behind proprietary formats that you used?
You now have an objection to someone using Excel? Menzie addressed your latest stupid question. Move on dude – perhaps to another blog.
rsm: Show my work? I took the point estimate and I took the reported standard error, and I added and subtracted 1.96 times the standard error. There is no “work” to be shown, unless you want me to show simple addition and subtraction.
You ask why the standard errors are so small. I do wonder, do you know anything about statistics? You can read the documentation I provided (I hope you can read).
You might as well put standard errors on your hair color. That would be equally useful – although more difficult to calculate.
Standard error for hair color is simple: dye greying hair and hope nobody notices.
What about the standard error for receding hairline?? (asking for a friend of a friend of a friend’s uncle)
USGS just came out with a new Bakken survey.
https://www.usgs.gov/news/national-news-release/usgs-releases-oil-and-gas-assessment-bakken-and-three-forks-formations
Previously they had looked at it in 2008 and 2013. First two studies had a big growth, from 4B BO to 8B BO. (Mostly from adding in Three Forks stratum, ignored in 2008). The more recent study is down to 4B BO. But this is basically explained by production (~3B BO) during the time frame, as well as increases in proved reserves (donno exactly, guessing that’s the other B). (USGS excludes proved reserves and prior production.)
Little surprised they didn’t increase their number. But it seems like their 2013 study was basically valid. Peak oilers who said it was crazy high were wrong (really just the cumulative production is already more than what they thought would happen.) But also some of those thinking it was too low (e.g. Harold Hamm said it should be 24B BO) were wrong too.
Really wish they did a full resolution of production between the two studies and proved reserve changes. And also just a bit more discussion on what they learned versus old study. But they sort of just present it and leave that to people to download and compare the recent one with the 2008 one.
Off-topic: National security, insurrectionists, conspirators of murder
https://www.politico.com/news/2021/12/18/stop-the-steal-founder-jan-6-committee-gop-lawmakers-525345
In today’s episode of “Chuckle To Yourself” news, we pick up where we last left off with our hero, the alter ego of Uncle Moses~~”Biffy The Slipseater”. He’s reading month old hardcopies of NYT and laughing out-loud in an insane Joker (of Batman fame) manner, with no one in earshot. In-between insane laughs he keeps repeating the words “shortages”…… “respect”……. and “compensate” and envisioning in his mind’s eye a vacuous space of nothingness in meaning…….
https://www.nytimes.com/2021/11/09/us/politics/trucker-shortage-supply-chain.html