Prediction results for september

I have discovered the tool of Yuri Arcurs that can predict your monthly earnings based on the sales you had so far. This tool uses historical data of Arcurs. I started to use the tool and compare it to the linear prediction in my previous blog entry here. Now that I have reached the end of the first full month with logged predictions, here are the basic statistical details:

	Arcurs's prediction	Linear prediction
Median	$63.78	$61.6
Average	$69.17	$68.06
Minimum	$57.71	$54.83
Maximum	$111.44	$123.8
Standard deviation	$12.53	$16.01

As you can see there are not too many changes in the data if you compare them to the previously published ones (E.g. minimum and maximum values did not change, they were all reached by the 10th of the month). It still shows that the linear prediction is not that different.

Below you can also see the chart showing the relations between the current earnings and the two predictions.

As you can see I have stopped collecting the data after 27th of Sept, because I don't think anyone would need any prediction on their earnings so late in the month.

Here is how close the two predictions get to the real number ($55.91):

	Arcurs's prediction	Linear prediction
Median difference	+$7.87	$5.69
Average difference	+$13.26	$12.15
Closest match	+$1.8	-$0.05
Maximum difference	+$55.53	$67.89

So on the first 5 days of the month Arcurs's prediction gives more precise numbers (+86% overprediction instead of the linear's 106%). However this advantage disappears soon in the last 7 days of my measurements there were several predictions over $7 from Arcurs's tool meanwhile the linear prediction have had less than $5 as its error. The smallest error in the last seven days was $0.34 (or 0.5%) with the linear prediction and $1.91 ( or 3.2%) with Arcur's tool. So if you are as eager as I am for new numbers (and your portfolio's profile is similar to mine or has small amount of sales during a day), than I suggest you to use Arcurs's tool in the early days. Arcurs were predicting consequently closer numbers to the final result - the 10 predictions I made during the first 5 days have had more than $11 advantage for Arcurs on average. Then after 5-7 days you can switch to the linear prediction algorithm. At least that's what I will do till my earnings don't start to get different. (e.g. if my total earnings reach several hundreds of dollars a month).