I have a large excel file that has monthly sales per customer for January - December 2016. I want to predict what their sales will be in January 2017.
You could average each client's data and ignore the zeros with a formula like
=AVERAGEIF(D2:D12,"<>0)
D2:D12 would be the range of a single client's sales variable and it would give you a monthly average for that client that you could use for January Predicted Sales.
You have several problems to solve:
Determining (a) candidate forecasting model(s) to use.
Organising your existing data to test whether such model(s) are actually suitable, performing such tests and selecting (a) suitable model(s) [There may be more than one model to be used dependent on whether your data are homogeneous or not.]
Organising your existing data to apply your chosen model(s) for the
purposes of making your prediction. (A different organisation to 2. may be required.)
Your description talks about "sales" but the data sample you provided mentions "claims". These are very different entities - sales (dependent on what type of sales) may well be as frequent as monthly, but claims are likely to be a lot less frequent. If this is the case and claims are highly infrequent, then there is little sense in trying to predict an individual customer's claim. In such a case it would make more sense to predict the aggregate level of claims across a group of customers.
With all modelling, and particularly with forecasting models, context is highly important in steering towards which particular types of model are likely to be suitable. As it is, you have provided no context about what your data really represents, so are unlikely (beyond random chance) to find that any solution offered to you is actually going to be suitable. A solution might compute but, in the context in which you are operating, will it provide anything like a sensible or justifiable set of forecasts?
The "AverageIf" solution may be sufficient; however, you may be able to do better if there is in fact any trends/seasonality in the data that could be used to modeling advantage. For each customer, I would check for autocorrelation in the data. "Autocorrelation, also known as serial correlation, is the correlation of a signal with a delayed copy of itself as a function of delay. Informally, it is the similarity between observations as a function of the time lag between them."(https://en.wikipedia.org/wiki/Autocorrelation) For instance, if there is significant autocorrelation at lag = 12, this would suggest yearly seasonality in the data (maybe every January is similar). There is a nice tutorial to analyze autocorrelation in Excel at:
http://www.real-statistics.com/time-series-analysis/stochastic-processes/autocorrelation-function/
If autocorrelation does exist, it would likely then be useful to perform regression with that time component(s). If there is a trend with time in additional to a cyclical component, that should also be factored into the regression (i.e., such as a "Year" variable); or a more sophisticated time series method could be applied that would accomodate trend and autocorrelation such as an Autoregressive Integrated Moving Average (ARIMA) model:
https://en.wikipedia.org/wiki/Autoregressive_integrated_moving_average
Excel has a forecasting function that might help:
FORECAST.ETS function
Calculates or predicts a future value based on existing (historical) values by using the AAA version of the Exponential Smoothing (ETS) algorithm. The predicted value is a continuation of the historical values in the specified target date, which should be a continuation of the timeline. You can use this function to predict future sales, inventory requirements, or consumer trends.
This function requires the timeline to be organized with a constant step between the different points. For example, that could be a monthly timeline with values on the 1st of every month, a yearly timeline, or a timeline of numerical indices. For this type of timeline, it’s very useful to aggregate raw detailed data before you apply the forecast, which produces more accurate forecast results as well.
Syntax
FORECAST.ETS(target_date, values, timeline, [seasonality], [data_completion], [aggregation])
And you can see it in action in a workbook from the FORECAST.ETS.SEASONALITY page:
Download a sample workbook
Related
An example, the time someone left home and the time someone called 9-1-1 and put these points in to predict ideally the time of incident on an excel format. I can put in a time in column a and column b but all it does is give me the half way point between the two. example column a says 12:00 and column b says 1:00 and the result would be 12:30. If I can get some thing more predictive using this approach, that is ideally what I'm looking for.
I used some of the standard functions in Excel to predict time based series.
We were looking at predicting data points for 1mis, 3mis and 6mis (mis = Months In Service).
We found that the forecast() function with some "fiddle" factors - sorry finely tuned polynomial assumptions - gave a reasonable prediction for our needs. We fed it steps of historical data to see the performance until it was suitable for what we needed.
Im trying to build an excel sheet that calculates synthetic options prices and greeks for time series data to model intraday options pricing, input is simply intraday price data, say Tick level to 5 minute interval. I found this https://www.thebiccountant.com/2021/12/28/black-scholes-option-pricing-with-power-query-in-power-bi/ which provides for powerBI and Black Scholes but possibly not very accurately. I prefer the Binomial method (I have used this excellent tutuorial to build a manual version for a large number of strikes but it takes a long time to calculate and is very very complex and also inaccurate due to not being able to calculate many steps before topping excel out: https://www.macroption.com/binomial-option-pricing-excel/).
Does anyone have any idea if this is possible to create an entire column in Power Query that will calculate bionomially derived options pricing using >100 even up to 1000 steps? The reason is intraday pricing using high resolution data 5min, 1min, Seconds and Tick I think needs a large number of steps to properly converge. This is just about doing a good enough model that can be used for visualising the progress of a trade on a given day.
Any pointers on how this could be done and calculated using M Language would be much appreciated and useful!
I work for a hospital that is part of a larger network. We were recently asked by our corporate overlords to address the use of a specific laboratory test. in general, this test should only be performed daily, which should be considered to corresponded to a 24 hour period from last draw. sometimes, however, based on when people arrive to the hospital (e.g. 7pm), and in the interest of bundling labs for a single draw, they may be drawn sooner to coincide with routine testing i.e. 5am. it would never be necessary to otherwise need to repeat within a short (8 hour) window, particularly on the same day.
we have been asked to validate to see if we are adhering to this general practice, as testing any more frequent than that, say, within 12h of a previous test, has no real clinical value and thus adds unnecessary cost.
To address this issue I was given a dataset that among other items includes all instances the lab was performed including collection date and time.
please see HIPPA-safe example below (to be clear, no real data and identifiers are not real); the actual dataset has over 4,174 entries corresponding to 1,328 unique persons. everyone had at least one test performed, not everyone had >1.
I THINK what I want to do is an IF formula that reads the antecedent cell to 1) check if same person and 2) if so, perform a subtraction of the time stamp to display the relevant difference in time, which I can then filter, create histogram, etc. does this seem like a reasonable approach? is there a more preferable method to facilitate analysis? do any other forms of analysis come to mind?
=IF(B2=B1, D2-D1, "n/a")
example data set with formula:
any other forms of analysis come to mind?
By the looks of it you should consider taking the values under "Results" into account, assuming there is a band that might be considered 'normal' readings. The "one in 24 hours is sufficient" rule of thumb may well be appropriate for a series of values within the 'normal' band but not so much so if readings are close to 'danger level'.
That is, in some cases a higher than 'standard' frequency of monitoring may be in the patient's interest, even if not hospital policy, so it may be worth separating the "less than 24 hours interval" readings into those where the higher frequency provided information of little value (eg readings remaining within a 'normal' band) from any that crossed into or out of the band and/or large changes in value. This though may be more a matter of statistical analysis than programming and depend upon whether any action might be taken as a result of such "extra" readings.
How Can I Model Multiple Short Time Series Samples?
For example, let's say I have a new subject each month, and I measure each subject every day for the entire month. I then want to model these multiple strings of independent time series because I assume that there is an underlying pattern that applies to all 12 subjects. However, a time series with an n of 30 is too short to model, so is there some way to group these 12 time series together for a parallel analysis?
I imagine the way to handle this is similar to how one might handle a time series with multiple breaks of unknown length. Unfortunately, I unaware of how to deal with this type of data structure.
Any thoughts on where to even begin? What terms I should research?
Well. Depends on what you're interested in. Makes it a lot easier if we know what kind of data you have, and what you're trying to analyse.
Trying to answer your question: If you assume that there is some underlying structure which is homogenous for, say, 6 of the subjects, and different for the other half, you can just pool the two data sets and do some kind of group-mean analysis. If you're interested in a temporal change over the 12 months, then you need to assume that each subject are homogenous across whatever variable you're measuring.
Normally, for e.g. timeseries in economics, what you're describing is called "censored" or "truncated data".
If we want to measure the income of everyone in a country, we do this by checking electronic paychecks or something. But some people at the end of each tail, may not have a visible income. Poor people may be earning income in other ways, and rich people may want to hide some of their income. This is censored data, and any advanced timeseries stats book will have something on that.
Truncated data is similar. Just imagine income again. If we truncate everyone who makes < 10,000$ a year, then this will "cut off the end" of your distribution. There are also remedies for this. Again check an advanced time series book.
Hope this helped a bit.
We have a application where users enter prices all day. These prices are recorded in a table with a timestamp and then used for producing charts of how the price has moved... Every now and then the user enters a price wrongly (eg. puts in a zero to many or to few) which somewhat ruins the chart (you get big spikes). We've even put in an extra confirmation dialogue if the price moves by more than 20% but this doesn't stop them entering wrong values...
What statistical method can I use to analyse the values before I chart them to exclude any values that are way different from the rest?
EDIT: To add some meat to the bone. Say the prices are share prices (they are not but they behave in the same way). You could see prices moving significantly up or down during the day. On an average day we record about 150 prices and sometimes one or two are way wrong. Other times they are all good...
Calculate and track the standard deviation for a while. After you have a decent backlog, you can disregard the outliers by seeing how many standard deviations away they are from the mean. Even better, if you've got the time, you could use the info to do some naive Bayesian classification.
That's a great question but may lead to quite a bit of discussion as the answers could be very varied. It depends on
how much effort are you willing to put into this?
could some answers genuinely differ by +/-20% or whatever test you invent? so will there always be need for some human intervention?
and to invent a relevant test I'd need to know far more about the subject matter.
That being said the following are possible alternatives.
A simple test against the previous value (or mean/mode of previous 10 or 20 values) would be straight forward to implement
The next level of complexity would involve some statistical measurement of all values (or previous x values, or values of the last 3 months), a normal or Gaussian distribution would enable you to give each value a degree of certainty as to it being a mistake vs. accurate. This degree of certainty would typically be expressed as a percentage.
See http://en.wikipedia.org/wiki/Normal_distribution and http://en.wikipedia.org/wiki/Gaussian_function there are adequate links from these pages to help in programming these, also depending on the language you're using there are likely to be functions and/or plugins available to help with this
A more advanced method could be to have some sort of learning algorithm that could take other parameters into account (on top of the last x values) a learning algorithm could take the product type or manufacturer into account, for instance. Or even monitor the time of day or the user that has entered the figure. This options seems way over the top for what you need however, it would require a lot of work to code it and also to train the learning algorithm.
I think the second option is the correct one for you. Using standard deviation (a lot of languages contain a function for this) may be a simpler alternative, this is simply a measure of how far the value has deviated from the mean of x previous values, I'd put the standard deviation option somewhere between option 1 and 2
You could measure the standard deviation in your existing population and exclude those that are greater than 1 or 2 standard deviations from the mean?
It's going to depend on what your data looks like to give a more precise answer...
Or graph a moving average of prices instead of the actual prices.
Quoting from here:
Statisticians have devised several methods for detecting outliers. All the methods first quantify how far the outlier is from the other values. This can be the difference between the outlier and the mean of all points, the difference between the outlier and the mean of the remaining values, or the difference between the outlier and the next closest value. Next, standardize this value by dividing by some measure of scatter, such as the SD of all values, the SD of the remaining values, or the range of the data. Finally, compute a P value answering this question: If all the values were really sampled from a Gaussian population, what is the chance of randomly obtaining an outlier so far from the other values? If the P value is small, you conclude that the deviation of the outlier from the other values is statistically significant.
Google is your friend, you know. ;)
For your specific question of plotting, and your specific scenario of an average of 1-2 errors per day out of 150, the simplest thing might be to plot trimmed means, or the range of the middle 95% of values, or something like that. It really depends on what value you want out of the plot.
If you are really concerned with the true max and true of a day's prices, then you have to deal with the outliers as outliers, and properly exclude them, probably using one of the outlier tests previously proposed ( data point is x% more than next point, or the last n points, or more than 5 standard deviations away from the daily mean). Another approach is to view what happens after the outlier. If it is an outlier, then it will have a sharp upturn followed by a sharp downturn.
If however you care about overall trend, plotting daily trimmed mean, median, 5% and 95% percentiles will portray history well.
Choose your display methods and how much outlier detection you need to do based on the analysis question. If you care about medians or percentiles, they're probably irrelevant.