# Lake County Sauvignon Blanc Price Forecast

This post presents my forecast for Lake County Sauvignon Blanc prices for 2016. It’s of very limited use this year for two reasons. First, I doubt there is much unsold fruit out there right now, if any. Second, Lake County is a small enough market that the some of the guys on the front lines can estimate a price pretty accurately with no or simple quantitative methods. This forecast, however, is purely quantitative, so it serves as a good check of my forecasting methods and may help some people who are looking ahead to 2017 to craft their pricing strategy. It only uses information available since April. If you are interested in this variety or grape price forecasting in general, I would encourage you to read all of this post, not just the price prediction. The confidence intervals are important and learning a bit about the accuracy and methodology could provide useful context.

The forecast indicates a virtually 100% chance that prices will be higher than they were last harvest, dependent upon inflation. It is a bold and aggressive forecast and one that should offer cheer to the sometimes underappreciated winegrowers of Lake County. And, I am proud to say that my intern, Jacob Tehrani did a great job and, frankly, did most of the work.

**Forecast**

The forecast price for 2016 is **$1,303.50** per ton. This assumes a **2.1%** consumer price index increase for 2016. If inflation differs, the price should shift proportionately.

**Precision / Confidence Intervals**

The graphic below presents the confidence intervals and is followed by a plain English explanation of what that means.

There is a

**60%**chance that the price will be between**$1,271.36**and**$1,335.63**, indicating a**20%**the price will be below the lower bound and a**20%**chance it will be above the upper bound.There is a

**80%**chance that the price will be between**$1,254.56**and**$1,352.43**, indicating a**10%**the price will be below the lower bound and a**10%**chance it will be above the upper bound.There is a

**90%**chance that the price will be between**$1,240.69**and**$1,366.31**, indicating a**5%**the price will be below the lower bound and a**5%**chance it will be above the upper bound.There is a

**95%**chance that the price will be between**$1,228.66**and**$1,378.34**, indicating a**2.5%**the price will be below the lower bound and a**2.5%**chance it will be above the upper bound.There is a

**99%**chance that the price will be between**$1,205.14**and**$1,401.85**, indicating a**0.5%**the price will be below the lower bound and a**0.5%**chance it will be above the upper bound.

**Variables**

This model is based on multiple regression analysis. It uses four variables, plus an intercept. Below is an explanation of the variables, along with the P-values. A P-value is one measure of a variables validity. The lower it is, the better. If it is below .05 it is typically considered statistically significant in academia. However, in a marketing department, tolerances tend to be much higher, say .20.

**Trend:** This variables measures price growth over time. The **P-value is** **.0002**.
**General Grape Market Variable:** Incorporates the market for premium Sauvignon Blanc, in general. **P-value is** **.0010**.
**Supply Variable: ** Measures available supply. P-value <.0001.
**Macroeconomic Variable:** Incorporates the general economy into the model. **P-value of .0124**.

The P-values are all well within the threshold of acceptability.

**Intercept:** The intercept is just a mathematical component of the model's output and has a **P-value of .0072.**

**Accuracy**

The R-squared and P-values given are common measures of a model’s strength and they are the most important to me. However, I do use a few other measures that are easier to understand for many people. First of all, there is the chance that the whole model has some issue that makes it unusable – that it will shatter upon contact with reality. My estimate is that **there is a 2% chance that the model does not work**.

On average, the model’s prediction of previous years is off by 0.09%. Of its predictions for the past 20 years, the worst was high by 8%. The best was within 0.5%. Eight times it has been high by more than 0.5%; 11 times it has been low by more than 0.5%; and once it was within 0.5%. These results indicate the forecast is not inherently biased. In this case, bias does not mean that it has an opinion, just that it tends to predict either high or low.

The predictions have been within the expected confidence intervals as shown in the summary chart below:

**Final Thoughts**

Lake County has been the most challenging area to predict prices for of anywhere that I do try to predict. The model looks strong and quite precise, too. The confidence intervals are tight enough to make me nervous. After all, I am claiming that there is a 99% chance that prices will fall within 6% of my prediction. That’s good. I want to be precise. After all, if the confidence interval is too broad, then who needs the prediction in the first place? So, anyone agree or disagree with the forecast? Comment or get in touch.

Stay tuned for my Lake County Cabernet Sauvignon forecast, which should come out next week.

Listened to while writing this:

Matisyahu, *Live at Stubb's*