I have modeled the STEEM price scientifically, then forecasted it, and the results were shocking!

I don't use technical analysis to predict prices because it's pretty unreliable, and I have yet to find a scientific research that have proven it's accuracy over long term. I might use technical analysis sometimes for illustrative purposes, but if we want to be serious we need to use real mathematical tools.

I use Quantitative Analysis and Econometrical tools to model financial markets, just like any Ph.D. Quant professional on Wallstreet.

I have demonstrated the usefulness of quant models in my previous articles:

• https://steemit.com/bitcoin/@profitgenerator/let-s-calculate-the-probability-of-bitcoin-going-to-1000usd-and-above-pt-1
• https://steemit.com/bitcoin/@profitgenerator/let-s-calculate-the-probability-of-bitcoin-going-to-1000usd-and-above-pt-2
• https://steemit.com/steem/@profitgenerator/scientific-forecasting-of-sbd-volume

In this article I will forecast the STEEM price about 42 days in the future, and the results will be pretty dramatic.

---

Modeling STEEM

• I have downloaded the STEEM_BTC price from Poloniex with their API system, by calling:
  https://poloniex.com/public?command=returnChartData&currencyPair=BTC_STEEM&start=1&end=9999999999&period=900

• It is price data in 15 minute timeframe (lower timeframes have too much data, and high signal to noise ratio, and in some cases many null values, that distort stats severely, so I have chosen 15 minute). We have 4625 elements, until yesterday, spanning from Tue Jul 19 04:15:00 2016 UTC to Mon Sep 05 08:15:00 2016 UTC.

• I have used the good old ARIMA model to model the price, and we will achieve very low statistical error numbers.

• I converted the JSON file into CSV so we can work easily with the data.

• We look for a trend, which is obvious that there is one, if we just look at the chart, but we use objective tools to measure it, like unit root tests:

---

Augmented Dickey–Fuller Test:

type	asymptotic p-value	model
with constant	0.5833	(1-L)y = b0 + (a-1)*y(-1) + ... + e
with constant and trend	0.2408	(1-L)y = b0 + b1*t + (a-1)*y(-1) + ... + e
with constant and quadratic trend	0.135	(1-L)y = b0 + b1*t + b2*t^2 + (a-1)*y(-1) + ... + e

We observe that we have a constant at least and probably a trend , so we will need to differentiate having a d>0 in the ARIMA(p,d,q) model.

---

Kwiatkowski–Phillips–Schmidt–Shin Test:

T = 4625
Lag parameter = 70
Test statistic = 0.765265
P-value < .01

The KPSS test has an inverse null hypothesis so a p<0.01 also confirms that we need to differentiate.

---

• Then we look for the p and q values with a Partial autocorrelation function

We observe that the p and q values will probably be between 0 and 1. But I have tested all possible combinations from 0 to 4.

---

• Then I went through all combinations from 0 to 4, an surprisingly the best model is ARIMA(0,1,0) or I(1), technically we just have a constant and a differentiation of it.

Constant coefficient: −6.79381e-07
Constant standard error: 9.27125e-07 (very accurate)

• According to Wikipedia an ARIMA(0,1,0) with a constant resembles:

An ARIMA(0,1,0) with a constant, given by — which is a random walk with drift.

• The model fits perfectly (I have made 1 line wider to be more visible):

• The actual vs predicted chart shows that they are aligned very well, with a few exceptions:

QQ Plot:

---

Forecasting STEEM

I have forecasted the STEEM_BTC price to 4000 units in the future, which for 15 minute chart means 41.6(6) days.

So if our last data unit is Mon Sep 05 08:15:00 2016 UTC this means that our forecast will last until Mon Oct 17 00:15:00 2016 UTC.

The error numbers are ultra low (for the actual data of course), so this means that our forecast data should be very accurate:

Error	Number
Mean Error	-1.0082e-20
Mean Squared Error	3.9738e-09
Root Mean Squared Error	6.3038e-05
Mean Absolute Error	2.9055e-05
Mean Percentage Error	-0.012379
Mean Absolute Percentage Error	0.91349

---

And since the price can't go below zero perhaps a more correct chart would look like this:

Where the area below the line has an equal probability of happening with the area above the line. So the price should be inside the green area with a 95% probability.

We also observe that it's very unlikely that the price will go above 0.005 BTC, with the probability of <2.5% of that happening in the next 41 days.

---

This means that the STEEM price at Oct 17th should be :

0.00043 BTC

Last price is 0.00131450 BTC, so this would be a -67.28% drop in the next 41 days!!

---

So given that our model is pretty accurate with an absolute error of only 0.000029055 from the mean, I think we have some problems.

I don't want to scaremonger here, and I could be wrong, but if I am not, then we have a problem, and the community has to do something about it. I hope @dan and @ned is already aware of the instability of STEEM and are looking for solutions to stabilize STEEM.

The market cap has already went down a huge amount so a further decrease is not unlikely. We as a community need to do something about it, to make STEEM stronger and more trusted.

I will write my next article about how to stabilize the STEEM price and ideas for long term growth. Stay tuned!

---

Disclaimer: The information provided on this page might be incorrect. I am not responsible if you lose money using the information on this page! This is not an investment advice, just my opinion and analysis for educational purposes.

Title Image Credit: @d3m0t3x
Chart Image Credit: https://poloniex.com

---

Upvote & Follow Me: @profitgenerator