Down goes the Dow

the idiot wrote:Predictions are very difficult, especially about the future... (Neils Bohr, Yogi Berra)

And yet, here is (an update of my March 2018) prediction:

[IMG]

]

Based on past behaviour of the market, I "predicted" (expected is a better word) the index to stay within the two curves closest to the best fit trend with an 80% probability (i.e., those curves are the 10th and 90th percentile range of past fluctuations away from the best fit trend. If my 80% chance was right (impossible to know with only one trial), then if we rewound the clock and tried this again the index ought to stray higher or lower one time in five.

Here's what the "prediction" (expectation) made in March 2018 looked like out to about this time in 2023 (5 year look ahead from then):

10% chance of being somewhere between 1600 (possibly lower) and 2350

40% chance of being between 2350 and 3800

40% chance of being between 3800 and 5450

10% chance of being higher than 5450

Given that we have progressed more or less on the best fit trend for the year following the first "prediction," we should update the look ahead accordingly, and you can't see this in the graph, but my spreadsheet gives me (for 5 years ahead, or this time in 2024):

10% chance of being somewhere between 1800 (possibly lower) and 2650 (- 38% to -9%)

40% chance of being between 2650 and 4250 (-9% to +46%)

40% chance of being between 4250 and 6100 (+46% to +110%)

10% chance of being higher than 6100 (> +110%)

Your ranges are so wide that I'd hardly call this a good/predictive model. This is equivalent to "well it could go up, or maybe down, but also maybe stay in a nice range."

Racket wrote:Your ranges are so wide that I'd hardly call this a good/predictive model. This is equivalent to "well it could go up, or maybe down, but also maybe stay in a nice range."

It's not obvious to me you understand what the model is trying to convey. It is showing the expected future index values, within specific confidence bounds, based solely on an analysis of past results.

Anyone who claims to be able to make a tighter estimation with higher confidence doesn't really understand probability and statistics, and / or has not spent time with these price data and learning about their statistical properties. Daily price changes are not normally distributed; they have heavy (fat) tails on both sides (positive and negative price changes). Here's a crude representation of data since 1950:

The behaviour of the index is largely controlled by extreme price changes. If you remove the ten largest or ten smallest daily price changes since 1950 (over 17000 days...), you change today's index value by about a factor of two:

So, the reality is the wide confidence bounds are a necessary reality to capture the behaviour of the index.

If you know a better way to make more confident "predictions" I am all ears...

Yeah I honestly don't even know where to start with this word vomit.

If you want to know how to model stock prices then look at Ito calculus, geometric Brownian motion, and brush up on your partial stochastic differential equations. Whatever normal distribution you're using here is extraneous and means nothing.

Ghost of Igloi wrote:

I didn’t move the goalposts

2015 - Major downturn "SOON"

2016 - "

2017 - "

2018 - "

2019 - "

Racket wrote:Yeah I honestly don't even know where to start with this word vomit.

If you want to know how to model stock prices then look at Ito calculus, geometric Brownian motion, and brush up on your partial stochastic differential equations. Whatever normal distribution you're using here is extraneous and means nothing.

As my name acknowledges, I am an idiot, and claim nothing different. I am keen to see your better "prediction." I imagine we will all benefit from it.

the idiot wrote:As my name acknowledges, I am an idiot, and claim nothing different.

Let me elaborate... I don't think I'm an idiot in the broadest sense (I think I have at least average intelligence), and I have a decent education, earn a decent living giving technical advice to people, have a couple of advanced degrees and have published some peer reviewed journal papers. What I mean is more specific, in that I think I am an idiot in relation to financial markets, in spite of having spent a lot of my > 50 years thinking about them. I've come to believe that financial markets are almost entirely unpredictable, giving up no useful information about themselves, and having very little relationship with other information (except sometimes in hindsight, which is rarely useful).

You, and others in this thread may not be idiots in relation to the market, and may have learned something I have not about "predicting" future prices or index values. I have admitted defeat in that realm, and fall back on a purely statistical point of view, and a naïve, Bayesian one at that), with no belief or confidence in any deterministic or semi-deterministic methods.

I am a curious person, though, always looking to learn new stuff, so I will be delighted to have you elaborate you prediction methods for the group.

Racket wrote:

Ghost of Igloi wrote:

Calculator is thru 4/5/2019 so what I originally quoted is incorrect. However your and Seattle numbers are way off. Here is the math for pure S&P 500 Index return, not dividends reinvested.

Close 12/31/2017 = 2,743.15

Close 4/16/2019 = 2,907.06

The difference is 163.91. 163.91 / 2,743.15 = .05975247% (gross return, not annualized)

.05975247 / 471 (number of days since 12/31/2017) = .00012686 (daily return)

.00012686 x 365 = .04630499%

So S&P 500 Index Return January 2018 thru today 4.6%

Your and Seattle’s mistake, I suppose, was incorrectly using the first day’s open trade.

Igy

If you move that back by just a week to 12/24/2018 then the SP500 is up an insane amount. There's no denying that it's skyrocketed up in the first ever V recovery in history. A literal -20% from ATH bear market and out of that in like 2 months.

That would actually be moving it forward almost a year.

Mouth Breather wrote:

Racket wrote:

If you move that back by just a week to 12/24/2018 then the SP500 is up an insane amount. There's no denying that it's skyrocketed up in the first ever V recovery in history. A literal -20% from ATH bear market and out of that in like 2 months.

That would actually be moving it forward almost a year.

Yeah I misread it the first time and thought he was talking about January 1st 2019. The second half of my statement stands

Following up by calling out "racket" on this:

the idiot wrote:… I will be delighted to have you elaborate you prediction methods for the group.

Here is a test, if you are game. Give me your best estimate, of index prices for any five major stock indexes, three months from now by assigning high and low values that you consider to represent 80% confidence limits. We can check back in three months, and if your confidence limits are reasonable (neither overconfident or too narrow, nor underconfident and too wide), you will be right in 4 of 5 cases. If you get them all right, it will mean you were underconfident (set your limits too wide), and if you miss more than one it will mean you were overconfident (set your limits too narrow).

This is an accurate test of your (or anyone's) predictive capabilities. Are you game? Put up or shut up, they say... :-)

the idiot wrote:Give me your best estimate, of index prices for any five major stock indexes, three months from now by assigning high and low values that you consider to represent 80% confidence limits. We can check back in three months...

Or, if you like, do the same for any number of individual stocks, or any financial instruments you care to predict. A bigger number will give us a better read on your accuracy (20 predictions instead of 5, for example), but in any case, being wrong (actual value outside your 80% confidence limits) 20% of the time will be a perfect result.

Stock price movement is generally random so there is no deterministic model and a catch-all model that covers every possible scenario in a random walk isn't worth much. In my opinion, long term predictions are for fun and for bragging rights only so I don't make real predictions and when I do I'm usually wrong (but that's the life of options and futures traders like me).

Sorry if I was too hard on your model. At a minimum I think you'll need some sort of multivariate Gaussian process to collect stock price data, model it, and spit out a joint normalized distribution or something idk

And the timeframe isn't important either, pick 1 day, 1 week, whatever you like.

____

Word vomit >>> Crickets chirping...?

Racket wrote:...I don't make real predictions and when I do I'm usually wrong...

Ah OK, neither do I, which is why I called my graph a "prediction" not a PREDICTION.

The test I suggested is perfectly valid and it would be fun (and educational) to see how people manage.

Fact checker wrote:

Wait what? wrote:

Wow! It’s only April and it’s up 2.9%. This is great!

It’s actually 8.3% and that doesn’t include dividends.

Just another Igy lie. Add it to the long list.

mister wrote:

Ghost of Igloi wrote:

I didn’t move the goalposts

2015 - Major downturn "SOON"

2016 - "

2017 - "

2018 - "

2019 - "

+1

Sally Vix wrote:

mister wrote:

2015 - Major downturn "SOON"

2016 - "

2017 - "

2018 - "

2019 - "

+1

What do you think will be Igy's prediction for the market in 2020?

Sally Vix wrote:

Sally Vix wrote:

+1

What do you think will be Igy's prediction for the market in 2020?

Down 65%.

Mouth Breather wrote:

Sally Vix wrote:

What do you think will be Igy's prediction for the market in 2020?

Down 65%.

I just cashed out some of my mutual funds which returned 400% in the last 10 years. I will be sailing around the world. I guess I don't regret having invested in CDs.