Inflation+Prediction

This project explores different models designed to predict the future inflation rate based on historical economic information. ** Since the dependent variable in the data is a continuous variable, the model will be run for predictive purposes. ** ** We will use predictive multiple linear regression, predictive generalized linear regression, CHAID tree models, and Neural Networks to determine the most influential factors of our model and allow us to obtain a minimum 70 percent accuracy rating. **The data was collected from various sources of government and other reliable websites as well as professional articles, which helped us obtain monthly data of the U.S. expenditures and economic information from 1976 to 2011**. Based on our analysis, we see that the Consumer Price Index variable and Unemployment Rate variable are the most important predictive factors aligned with changes in the inflation rate. ** Our findings suggest that companies should pay attention to changes in the Consumer Price Index and unemployment rates over time when organizing their business plan. The value of the U.S. Dollar is ever changing, whether we as consumers know it or not. Since the United States left the gold standard in 1932, the relative value of the dollar has been both an indicator of political and economic stability and a key driver of consumer confidence. The decreasing value of the dollar (and therefore higher prices of consumer goods) is known as inflation and the rate at which it fluctuates is known as the inflation rate. While the U.S. government tries to control inflation via the Federal Reserve, its true value is dictated by the volatility of the market. The inflation rate has enormous consequences for the economy and investors. Negative effects of inflation include a decrease in the real value of money and other monetary items over time; uncertainty over future inflation may discourage investment and savings; and high inflation may lead to shortages of goods if consumers begin hoarding out of concern that prices will increase in the future. Positive effects include ensuring central banks can adjust nominal interest rates and encouraging investment in non-monetary capital projects. Companies use inflation rates when planning finances, a key step in optimizing one’s investments, cost management and financial needs. Thus, knowing which factors affect the inflation rate in order to make valid predictions about what future rates will be useful for businesses.
 * __EXECUTIVE SUMMARY:__**

**__PROCEDURE:__** The project will be divided into three steps. First, we will run a predictive model to identify what might be interpreted as causing inflation. Then, a cluster model will be used to identify other factors that accompany inflation. Finally, the results of the two types of analyses will be compared to come up with appropriate answers for the question above. ** We hypothesize that the inflation rate is dependent on several key economic factors. We also hypothesize that fluctuations in the inflation rate will be most strongly correlated with fluctuations in the CPI and Unemployment Rate, and fluctuations in several other variables will aid in the predictive ability of a statistical model. ** Based on the project plan, selected data was collected to support the model. With regards to the causal prediction, we see that the inflation rate depends in part upon exchange rates and the USD’s value relative to foreign currency. We can compare foreign inflation rates with that of the U.S. to assess stability of economies and possible financial investment strategies. Additionally, because the inflation rate weighs so heavily on foreign investment, it can have great implications for foreign investment, import, and export. Since both the UK’s and Japan’s economies have a strong impact on global financial market, their currency exchange rates (UK-Pound and Japanese-Yen) should be considered. We also selected the Unemployment-rate and Housing-rate of the U.S., which reflect the U.S. economic situation, to accompany in this predictive model. For the clustering model, we see ** that there are several key economic factors which affect the inflation rate. T **he most common measure of inflation rates is the Consumer Price Index (CPI), which is based on a survey of tens of thousands of items Americans commonly purchase. Our project will use other variables calculated by the Bureau of Labor Statistics to predict the inflation rate. These variables include ** Gasoline, Commodities, Miscellaneous, and Rubber and plastic. **Additionally, we will use data of average expenditures on Food and beverages,** Power and related products **, and ** Furniture and household durables **. The historical inflation rate is also listed for predictive purposes. We decided to use the monthly data from 1976 to 2011 for the project, since this period includes a part of the Cold War (1947-1991), the Asian economic crisis (1997-8) and the great economic crisis (2007-2010); thus, it can offer valuable insight into how and why the inflation rate changes.

**Completed Data Mining Steps:** ** Since the dependent variable in the data is a continuous variable, the model will be run for a prediction purpose. We will be using predictive linear regression, predictive generalized linear, CHAID tree models, and Neural Network models. Finally the best model of the three will be chosen as the predictive model. The choice will be made based on the standard error and predictive accuracy of each model. We also broke the data into two sections: those data pertinent to household and marketing purposes (CPI, Gasoline, etc.) and typically recognized as being indicators of consumer confidence, and those data more pertinent to financial and investment purposes (unemployment rate, exchange rates, etc.). **

**__MODELS__** ** First we compiled and cleaned the data, removing outliers and accounting for missing data.via analysis using data audit and transform nodes. ** ** We then separated the data based on whether it was deemed as “associative” or “causal,” using a filter node, and ran both through several models. We also partitioned the data in order to make sure our model will be predictive of data beyond that used to build it. **





** We examined each model nugget, and using a user input node could predict inflation rates based on acquired or predicted knowledge of our other variables. Below are results of predictor importance for our Neural Network using causal data. **



** This was our most accurate model. We also included an accuracy summary of this model, showing it was 90 percent accurate, well above our goal of 70 percent accuracy. Several other models had accuracy ratings of above 70 percent, but no model truly agrees on one single variable as the most important predictor. **



[T1] ** However, using an analysis node and comparing all the models together, we see that most of the models agree fairly well on their actual predictions. **

**__ DEPLOYMENT __** The associative model (using ‘Gasoline,’ ‘Commodities,’ ‘CPI,’ etc.) lends itself to use in the field of marketing. It represents the relationship between fluctuation in the inflation rate and price fluctuation of common household necessities. The causal model (using exchange rates, T-bill, and Moody’s Aaa rates, etc.) will be more useful in the financial sector. This model uses other historical financial data in order to predict the inflation rate, which has major implications for stock and investment trading, as well as global market trade. Both of these models require some considerable statistics acumen, and are mainly accessible only to those with a background in forecasting, econometrics, or intermediate statistics. Both neural net models are surprisingly accurate and could be used in market research to assess market prices and consumer confidence or in financial analysis assessing potential financial trading and/or global trading.

**__ LIMITATIONS __** ** We are very pleased with the accuracy of our models. It is important to realize that no single variable is in itself a good predictor of the inflation rate, however using several key variables together and examining their relationships with each other can create an accurate predictive model. A major contributor to the accuracy of our model is most likely the frequency of our data (monthly). Additionally, while our model is very accurate it is not over fitted and has the ability to still be accurately predictive when used with new data. While our model can give accurate predictions for the inflation rate, it still does not provide insight as to how companies should act considering the interest rate. ** ** The nature of our most accurate model, the Neural Network, is that while it can accurately predict an inflation rate given series of data, it is unclear how exactly the Neural Net comes to its conclusion. Dissimilar to linear regression or even CHAID models, we have no general rule and have to trust the hidden neurons of the Neural Net. Also, we should take into account the high correlation between our predictor data (i.e. exchange rates, unemployments rates, and other “causal” data) and the inflation rate itself. Especially in our linear regression model, we have to be aware of multicolinearity. ** ** Further analysis must also take into account intricacies of specific business employing our model. While our model accounts for previous historical spikes in the inflation rate, consumer prices, exchange rates, etc. it cannot predict extraordinary phenomena such as stagflation and economic depressions. **

**Resources:** Chang, Kuang-Liang, and Chi-Wei He. "Does the Magnitude of the Effect of Inflation Uncertainty on Output Growth Depend on the Level of Inflation?." //Manchester School// 78.2 (2010): 126-148. //EconLit with Full Text//. EBSCO. Web. 11 Apr. 2011. Berentsen, Aleksander, Guido Menzio, and Randall Wright. "Inflation and Unemployment in the Long Run." //American Economic Review// 101.1 (2011): 371-398. //EconLit with Full Text//. EBSCO. Web. 11 Apr. 2011. [] [] [] []