Guide to The Philanthropy Outlook Model 2019 & Marts & Lundy. Indiana University Lilly Family School of Philanthropy RESEARCHED AND

Transcription

1 Guide to The Philanthropy Outlook Model 2019 & 2020 P R E S E N T E D BY Marts & Lundy RESEARCHED AND W R I T T E N BY Indiana University Lilly Family School of Philanthropy JA N UA RY 2019

2 Variable Definitions and Sources Independent Variables 1 CONSUMER SENTIMENT (GCSENT) Consumer sentiment is an index computed based on monthly surveys covering personal finances, business conditions, and buying conditions. Data for consumer sentiment come from the Consumer Sentiment Index, Federal Reserve Bank of St. Louis (FRED), fred2/series/umcsent CORPORATE PROFITS (GCPROF) Corporate profits are corporate income after subtracting expenses. Data for corporate profits come from the Bureau of Economic Analysis, U.S. Department of Commerce, CORPORATE SAVING (GCSAVE) Corporate saving is corporate profits that are left over after taxes and dividend payments. Data for corporate saving come from the Bureau of Economic Analysis, U.S. Department of Commerce, B057RC1Q027SBEA#0 EMPLOYMENT (GEMP) Employment is a measure of the number of U.S. workers in the economy that excludes proprietors, private household employees, unpaid volunteers, farm employees, and the unincorporated self-employed. Data for employment from FRED, fred2/series/payems GROSS DOMESTIC PRODUCT (GGDP) Gross Domestic Product (GDP) is the value of the production of goods and services in the United States, adjusted for price changes, according to the Bureau of Economic Analysis, U.S. Department of Commerce. Data for GDP come from Table 1.1.5, Bureau of Economic Analysis, U.S. Department of Commerce, gov/itable/index_nipa.cfm HOUSEHOLD AND NONPROFIT NET WORTH (GNWORTH) Net worth for households and nonprofits is the net assets of households and nonprofits serving households after subtracting net liabilities. Data for the net worth of households and nonprofits come from FRED, research.stlouisfed.org/fred2/series/hnonwra027n TAX DUMMY (TAX DUM) The Tax Dummy is zero except for 1986 when its value is one and 1987 when its value is negative one. The Tax Reform Act of 1986 implemented a two-step change in the highest individual tax rate from 50% in 1986, to 38.5% in 1987, and then to 28% in One would expect a spike in giving in 1986 as households shifted their planned giving forward to take advantage of the higher marginal tax rate in Likewise, one would expect a trough in giving in 1988 once the new lower tax rates were in effect. The 1987 response could be positive or negative. In fact, the data show a large spike in 1986, followed by a substantial decline in 1987, and a return to normalcy in The explanation outlined here does not account for this behavior. Nevertheless, the effects are so large that we elected to model that behavior directly to avoid the effect of the one-time tax reform exerting an undue influence on the remaining coefficients. B

3 INDIVIDUAL/HOUSEHOLD ITEMIZERS AND NON-ITEMIZERS (GNITEM) Itemizers refer to those taxpayers who can itemize certain expenses on their household taxes, as opposed to taking the standard deduction. Data for the number of itemizers come from the Internal Revenue Service (IRS), Data for non-itemized giving come from the Philanthropy Panel Study, Indiana University Lilly Family School of Philanthropy, iupui.edu/research/current-research/philanthropy-panelstudy.html, and Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, givingusa.org INTEREST RATE FOR GOVERNMENTAL SECURITIES (DR1T) The interest rate for governmental securities is the rate of return on an asset after removing the effect of inflation. Data for the interest rates of governmental securities come from FRED, PERSONAL CONSUMPTION EXPENDITURES (GRCONS) Personal consumption is a measure of personal consumption expenditures, or goods and services purchased by U.S. residents. Data for personal consumption come from FRED, from stlouisfed.org/series/pcec96#0 PERSONAL CONSUMER EXPENDITURE INDEX 2 Personal consumer expenditures is the primary measure of consumer spending on goods and services in the U.S. economy. It accounts for about two-thirds of domestic final spending, and thus it is the primary engine that drives future economic growth. Personal consumer expenditures shows how much of the income earned by households is being spent on current consumption as opposed to how much is being saved for future consumption. 3 Data on consumer expenditures come from FRED, PERSONAL INCOME (GPINC) Personal income is the income received by persons from participation in production, government and business transfers, and government interest. Data for personal income come from Table 2.1, Bureau of Economic Analysis, U.S. Department of Commerce, gov/itable/index_nipa.cfm THE S&P 500 (GSP) The S&P 500 is the value of the Standard & Poor s 500 Index on December 31 of a given year. Data for the S&P 500 come from FRED, fred2/series/sp500 THE PHILANTHROPY OUTLOOK 2019 &

4 Dependent Variables GROWTH RATE FOR INDIVIDUAL/HOUSEHOLD GIVING (GIGIV) The growth rate for individual/household giving includes cash and non-cash donations to U.S. charities contributed by all U.S. individuals and households (including those who itemize their charitable contributions on their income taxes and those who do not). Historical data for the growth rate in individual/household giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, GROWTH RATE FOR FOUNDATION GIVING (GFGIV) The growth rate for foundation giving includes grants to U.S. charities contributed by all U.S foundations. Historical data for the growth rate in foundation giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, Foundation giving data in Giving USA are based on estimates produced by the Foundation Center ( and include grants from community, private (including family), and corporate foundations. GROWTH RATE FOR ESTATE GIVING (GBGIV) The growth rate for estate giving includes cash and noncash donations (bequests) to U.S. charities contributed by all U.S. estates (including those who itemize their charitable contributions on their estate taxes and those who do not). Historical data for the growth rate in estate giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, GROWTH RATE FOR CORPORATE GIVING (GCGIV) The growth rate for corporate giving includes cash and non-cash IRS itemized donations to U.S. charities contributed by all U.S. corporations and corporate foundations. Historical data for the growth rate in corporate giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, 2

5 GROWTH RATE FOR EDUCATION GIVING (GEDUCGIV) The growth rate for education giving includes cash and non-cash donations from itemizing and non-itemizing U.S. households to U.S. educational charities, including institutions of higher education, private K-12 schools, vocational schools, libraries, educational research and policy, and other types of organizations serving educational purposes. Historical data for the growth rate in education giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, GROWTH RATE FOR HEALTH GIVING (GHEALTHGIV) The growth rate for education giving includes cash and non-cash donations from itemizing and non-itemizing U.S. households to U.S. health charities, including nonprofit community health centers, hospitals, and nursing homes; organizations focused on the treatment and/or cure of specific diseases; emergency medical services; wellness and health promotion; mental healthcare; health research; and other types of health organizations. Historical data for the growth rate in health giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, givingusa.org GROWTH RATE FOR PUBLIC-SOCIETY BENEFIT GIVING (GPSBGIV) The growth rate for public-society benefit giving includes cash and non-cash donations from itemizing and nonitemizing U.S. households to U.S. public-society benefit charities, including independent research facilities, community development organizations, human and civil rights organizations, philanthropy associations, national donor-advised funds, United Ways, federated charities, and other types of organizations. Historical data for the growth rate in public-society benefit giving were derived from Giving USA 2018: The Annual Report on Philanthropy for the Year 2017, researched and written by the Indiana University Lilly Family School of Philanthropy and published by Giving USA Foundation, THE PHILANTHROPY OUTLOOK 2019 &

6 Stability of the Variables Used in the Forecast To estimate charitable giving in future years, we must generate estimates of the economic variables that affect giving. We can expect the accuracy of these estimates to be higher or lower based on each variable s historical variance. Deviations in the variables would affect our outlook for giving. The next section, Conditions That May Impact the Giving Predictions, explains the changes in the variables that would have to take place in order to change the outlook for giving for each source and the three subsectors included in this Outlook. CONSUMER SENTIMENT Consumer sentiment affects giving by corporations. This variable is generally an unstable economic indicator, meaning the likelihood that the growth rate for this variable will be considerably different than predicted is high. its predicted impact on giving by individuals/households, foundations, and estates is deemed highly reliable. 6 INDIVIDUAL/HOUSEHOLD ITEMIZERS AND NON-ITEMIZERS While this variable has influence on individual giving, it is an unstable indicator, particularly in the short-term given the changes in tax law. The likelihood that the growth rates for these variables will be considerably different than predicted is high. INTEREST RATE FOR GOVERNMENTAL SECURITIES The interest rate for governmental securities has significant influence on estate giving, in particular. This variable is a stable economic indicator. Therefore, its predicted impact on giving by estates is deemed highly reliable. Otherwise, this variable plays a small role in our predictions overall. CORPORATE SAVING AND CORPORATE PROFITS While these variables have significant influence on corporate giving, they are unstable economic indicators. The likelihood that the growth rates for these variables will be considerably different than predicted is high. PERSONAL CONSUMPTION Personal consumption affects giving to education. This variable is generally a stable economic indicator, meaning the projected growth rate is not likely to differ significantly from what is predicted in this Outlook. 7 EMPLOYMENT The employment rate is a stable indicator of giving, meaning the projected growth rate is not likely to differ significantly from what is predicted in this Outlook. Therefore, its predicted impact on giving by corporations is deemed highly reliable. 4 GDP GDP is generally a stable indicator of giving, meaning the projected growth rate is not likely to differ significantly from what is predicted in this Outlook. Therefore, its predicted impact on giving by foundations and corporations is deemed highly reliable. 5 However, GDP may fall if the U.S. economic environment experiences an exogenous shock as a result of recession, disaster, war, or other severe situations. HOUSEHOLD AND NONPROFIT NET WORTH Household and nonprofit net worth is a stable indicator of giving, meaning the projected growth rate is not likely to differ significantly from what is predicted in this Outlook. Therefore, PERSONAL CONSUMER EXPENDITURES Personal consumer expenditures affect giving to education, health, and public-society benefit. There are many different types of personal consumer expenditures, and the majority are stable economic indicators. This means that for most of these indicators, the projected growth rates are not likely to differ significantly from what is predicted in this Outlook. PERSONAL INCOME Personal income is a stable indicator of giving, meaning the projected growth rate is not likely to differ significantly from what is predicted in this Outlook. Therefore, its predicted impact on giving by individuals/households is deemed highly reliable. 8 THE S&P 500 While S&P 500 has significant influence on corporate, individual/household, and foundation giving, this variable is an unstable economic indicator. The likelihood that the growth rate for this variable will be considerably different than predicted is high

7 Predicting Giving for the Sources of Philanthropy This Philanthropy Outlook was constructed using econometric methods. We began building the prediction models by testing economic variables with established links to charitable giving. Specifically, we tested those variables that measure giving capacity and the cost of giving as reflected in tax rates. For each source of giving individual/household, foundation, estate, and corporate we selected the macroeconomic variables that best accounted for the growth rates in giving. There are a large number of variables that can potentially account for giving. It is not practical to include all the potential explanatory variables in each of the models. To identify those variables that best explain and predict giving behavior, we employed a two-step process. In the first step, we tested each possible combination of explanatory variables. From the results, those combinations of variables with the greatest explanatory power were identified. Then, the models with the highest explanatory power were re-estimated through 2005 and used to predict the remainder of the sample. The best model was the one that yielded predictions that most closely resembled actual giving behavior. The number of estimations required to employ this strategy depends upon the number of potential explanatory variables. As this number rises, the number of estimations increases dramatically. For instance, with five potential variables, 32 estimations are required; but with 10 potential variables, 1,024 estimations are required. Our experience suggests that 17 variables (131,072 combinations) is a practical limit. If we consider only variables that occur in the same year as the giving variable, this restriction is not serious. But, experience also suggests that variables from the previous year ( lagged variables ) may also be important. For that reason, the best model was selected in three steps: (Model Process 1) Only same-year variables were considered. The best model produced in this step was called the base model. (Model Process 2) Previous-year (lagged) variables were added to the base model, and the best model was chosen using the criteria described above. The result was referred to as the revised model. (Model Process 3) The revised model was tested to determine if the current-year variables contained in the model were still relevant once the lagged variables introduced in Model Process 2 were included. Additionally, if any variable had previously been included in the final model from the prior year but was not in the revised model, it was included here for testing. Very few variables from the base model were eliminated in this step. The result was called the final model. Predicting giving requires predictions of the explanatory variables used in the prediction models. Due in part to passage of the Tax Cuts and Jobs Act (TCJA) in December 2017, which will have economic effects beyond our model s ability to predict, we partnered with the University of Pennsylvania Wharton School of Business to include predictions for select economic variables from its Penn Wharton Budget Model in our model. These variables included GDP, number of itemizers, GDP deflator, employment, one-year treasury rate, and the Consumer Price Index. The Wharton School of Business specializes in predicting economic variables and the Penn Wharton Budget Model accounts for policy changes, including the TCJA. We began with an aggregate model that included growth rates in real GDP, consumer sentiment, employment, the GDP price index, and change in the one-year real interest rate. The growth rate in the monetary base was also included to capture monetary policy. Each of these six variables was regressed on a one-year lag. This model was self-contained in that it yielded predictions for the previously unpredicted variables through Most of the remaining variables were grouped into blocks and regressed on the six macroeconomic variables along with the lags. Once we predicted growth rates for the four components of giving, we were able to extrapolate the corresponding levels. Summing the four levels together yielded our prediction for the level of total giving. Then, we calculated the implied prediction for the growth rate of total giving. THE PHILANTHROPY OUTLOOK 2019 &

8 PREDICTING GIVING FOR THE RECIPIENT SUBSECTORS For predicting giving to the recipient subsectors education, health, and public-society benefit we used a similar process to the one described above for predicting the sources of giving. We began this process by using the same independent variables as the earlier models. We then tested additional macroeconomic variables that have either established or theoretical links to giving. Those variables that demonstrated both high explanatory power for subsector giving and were highly correlative with actual historical giving trends were selected for the subsector models. PREDICTING GIVING UNDER THE ALTERNATE SCENARIOS Along with the baseline predictions for the economic variables that we received from the Wharton School of Business, we also used its 90 th percentile and 10 th percentile projections for GDP to form the upper and lowers bounds of the giving projections that appear in Figures 1 and 2 in the Scenario Analysis section of this Outlook. Using these values to predict the other variables allowed us to approximately quantify how giving might vary under a better-than or worse-than expected economy. Table 1 GIVING PROJECTIONS BASED ON 10 TH PERCENTILE GDP ESTIMATE Table 2 GIVING PROJECTIONS BASED ON 90 TH PERCENTILE GDP ESTIMATE TOTAL 3.1% 3.5% TOTAL 4.9% 4.7% INDIVIDUALS 1.4% 2.3% INDIVIDUALS 4.2% 4.7% FOUNDATIONS 6.6% 5.5% FOUNDATIONS 7.8% 5.6% BEQUESTS % 8.0% BEQUESTS % 4.0% CORPORATIONS 1.8% 1.9% CORPORATIONS 4.2% 3.0% EDUCATION 1.5% 2.7% EDUCATION 7.4% 7.2% HEALTH 1.4% -0.2% HEALTH 8.9% 7.0% PUBLIC-SOCIETY BENEFIT 2.1% 3.3% PUBLIC-SOCIETY BENEFIT 3.8% 4.6% The giving variables predicted within The Philanthropy Outlook 2019 & 2020 are listed in Table 3a. Candidate variables used to model each source of giving are listed in Table 3b. Tables 4 and 5 provide the regression equations used to predict each giving type within The Philanthropy Outlook 2019 & Table 6 provides the ratio of the root-mean-squared error to the standard deviation for each giving type, and Table 7 displays summary statistics for the giving variables and explanatory variables used in the models. Figure 1 shows actual versus predicted growth rates for total giving for the years 2006 to Table 3a GIVING VARIABLES MODELED BY THE REGRESSIONS Dependent Variables GROWTH RATE OF NATIONAL INDIVIDUAL/HOUSEHOLD GIVING Name GIGIV GROWTH RATE OF NATIONAL CORPORATE GIVING GCGIV GROWTH RATE OF NATIONAL FOUNDATION GIVING GFGIV GROWTH RATE OF NATIONAL BEQUEST GIVING GBGIV GROWTH RATE OF NATIONAL EDUCATIONAL GIVING GEDUCGIV GROWTH RATE OF NATIONAL HEALTH GIVING GHEALTHGIV GROWTH RATE OF NATIONAL PUBLIC-SOCIETY BENEFIT GIVING GPSBGIV 6 6

9 Table 3b CANDIDATE VARIABLES USED TO MODEL EACH TYPE OF GIVING (Variables are in year-to-year rates of growth) Candidate Independent Variables Name * gigiv gcgiv gfgiv gbgiv geducgiv ghealthgiv gpsbgiv CONSUMER SENTIMENT (INDEX) GCSENT I I I CORPORATE PROFITS GCPROF I CORPORATE SAVING GCSAVE I CORPORATE TAX RATE DCTAX I DISPOSABLE PERSONAL INCOME GDPINC I EMPLOYMENT GEMP GDP GGDP I I I I HOUSEHOLD AND NONPROFIT NET WORTH GNWORTH I I I I I DUMMY VARIABLE FOR THE 1986 TAX REFORM NUMBER OF INDIVIDUAL/ HOUSEHOLD TAX ITEMIZERS TAXDUM I I I I GNITEM I I INDIVIDUAL/HOUSEHOLD TAX RATE DPTAX I INTEREST RATE FOR GOVERNMENTAL SECURITIES DR1T I MONETARY BASE GMBASE PERCENT HEALTHCARE CONTRIBUTED TO GDP PCTCONTTOGDPPCEHEALTHCARE PERSONAL CONSUMPTION GCON PERSONAL CONSUMER EXPENDITURES: CLOTHING GCLOTHING I COMMUNITY SCHOOL SERVICES GCOMMUNITYSCHOOLS I EDUCATION GEDUCATION EDUCATION (HIGHER) GSERVICESHIGHERED EDUCATION (PREK-12) GNURSERYTOHS I EDUCATION SERVICES GEDUCSERVICES I I FOREIGN TRAVEL GFOREIGNTRAVEL I FURNISHINGS GFURNISHINGS GOODS: JEWELRY AND WATCHES GGOODSJEWELRYANDWATCHES GOODS: MOTOR VEHICLES GGDOODSMOTORVEHICLES GOODS: NEW MOTOR VEHICLES GGOODSNEWMOTORVEHICLES GOODS: TEXTBOOKS GGOODSEDUCBOOKS HEALTH GHEALTH I HEALTHCARE SERVICES GHEALTHCARESERVICES I NONPROFIT SALES GNPORECEIPTSALES I NONPROFIT SERVICES GNETNPOSERVICES I NONPROFIT FINEXP SERVICES GSERVICESFINEXPNPO PHARMACEUTICALS GPHARMA I I RECREATION GREC I I RECREATION SERVICES GRECSERVICES I SOCIAL AND RELIGIOUS SERVICES GSOCIALSERVICESANDRELIG I PERSONAL GIVING GPGIV I PERSONAL INCOME GPINC I PERSONAL SAVING GPSAVE I I PERSONAL SAVING RATE DPSRATE I I I PREVIOUS YEAR S VALUE OF THE GIVING VARIABLE PROPORTION OF MONTHS IN WHICH THE ECONOMY WAS IN A RECESSION I I I DRECM I S&P 500 (INDEX) GSP I I I I TOTAL GIVING GTGIV I I * The second column contains the names of the variables that appear in one or more of the models. See Tables 4 and 5 for the final models. : Either the current or lagged value of this variable is included in the final model. : This variable was tested for inclusion in the final model but was rejected. Empty cells reflect variables that were not tested within the specific giving model.

10 Table 4 MODELS FOR PREDICTING GIVING BY DONOR TYPE AND STEP-AHEAD ACCURACY CHECK GIVING BY INDIVIDUALS/HOUSEHOLDS gigiv = gpinc gnworth taxdum gcsent gnitem dptax dpsrate 0.164grpsave 0.090gcsent -1 Adjusted R 2 =0.640, Sample: (n=63) GIVING BY CORPORATIONS gcgiv = gcsave gcsent dctax taxdum ggdp 0.429gcprof gcsent -1 Adjusted R 2 =0.386, Sample: (n=63) GIVING BY FOUNDATIONS gfgiv = gsp 0.566gnworth gsp gcsent ggdp e t e t 2 Adjusted R 2 =0.467, Sample: (n=63) GIVING BY ESTATES gbgiv = gsp 1.387dr1t 1.741gnworth gnworth e 1 Adjusted R 2 =0.252, Sample: (n=63) Notes: e 1 and e 2 are one-period and two-period lagged residuals from the respective models. These models use 2005 as the first prediction. Table 5 MODELS FOR PREDICTING GIVING TO THE RECIPIENT SUBSECTORS AND STEP-AHEAD ACCURACY CHECK EDUCATION GIVING geducgiv = gsp 1.510grdpinc dpsrate ggdp taxdum grnitem gpgiv 3.101NetNPOServices gHealth gEducServices NetNPOServices gHealth gRec gSocServAndRelig CommunitySchools 1 Adjusted R 2 =0.765, Sample: (n=57) HEALTH GIVING ghealthgiv = ggdp gnworth 0.741gtgiv gHealthcareServ gNurseryToHS 2.762gClothing 7.351gNPOReceiptSales 0.939gPharma drecm gEducServ gNPOReceiptSales -1 Adjusted R 2 =0.462, Sample: (n=57) PUBLIC-SOCIETY BENEFIT GIVING gpsbgiv = gnworth taxdum gForeignTravel gTotGiv 0.292gsp dpsrate grpsave gForeignTravel gPharma gRecServices gRec e t 1 Adjusted R 2 =0.673, Sample: (n=62) Notes: e 1 is a one-period lagged residual from the respective models. These models use 2005 as the first prediction. 8

11 Table 4 includes the dependent variables for giving by individual/households, corporations, foundations, and estates. Table 5 includes the dependent variables for giving to education, health, and public-society benefit. These variables are on the left side of the equation. On the right side of the equations within Tables 4 and 5 are the independent variables that comprise each model, along with their coefficients and a constant variable. Most of these variables are in the form of growth rates and are therefore percentages. The -1 subscript after a variable name refers to the prior year s value. For instance, gsp 1 in the individual/household equation for the year 2019 is the growth rate of the S&P 500 in We can also use the equation for individual/household giving as an example for how the results are interpreted. This equation says that a 1% increase in the growth rate for household net worth (gnworth) is associated with a 0.273% increase in the growth rate for personal giving. These effects are summed for each variable, as well as for the constant, which gives us our predicted growth rate for that year. The abbreviations for each of the variables are listed in Tables 3a and 3b. The R 2 value below each equation is a measure of how well the model explains the results upon which it is based. R 2 values can range from 0 to 1, with higher values indicating greater explanatory power. For instance, the adjusted R 2 of for giving by individuals/households means that the model accounts for 64.0% of the variance in the growth rate for this series. In general, the R 2 s reported above are satisfactory, and in some cases superior, given the typical difficulty in explaining growth rates. The ability to explain historical behavior need not translate into high-quality predictions of future growth rates. While R 2 values are reported here, they were not the criterion used for model selection. Instead, models were selected based on a combination of root-mean-squared error (RMSE), Bayesian or Schwarz information criterion, as well as the Akaike information criterion. R 2 values are reported for their ease of interpretation and ubiquity. The sample identifies the years included in the data series used to estimate the models. THE PHILANTHROPY OUTLOOK 2019 &

12 Prediction Quality A common check of prediction quality is to re-estimate the model, setting aside the most recent observations. The revised model is then used to produce predictions over the set-aside observations. These predictions can be compared directly to their corresponding actual values. Here, we reestimate the models using data through These models are then used to construct step-ahead values year by year through the end of the sample. Step-ahead analyses assume that the prior year s values are known for generating the current year s value. In the table below, RMSE is the root-mean-squared error defined as: Where Actual is the actual growth rate, Prediction is the predicted growth rate, and T is the number of years in the prediction period, standard deviation (Std.Dev) is defined as: Average is the average of the actual growth rates. Table 6 RATIO OF THE ROOT-MEAN-SQUARED ERROR TO THE STANDARD DEVIATION FOR EACH GIVING TYPE The third column in the table contains the ratio of the RMSE to the standard deviation. If the predicted values are no better than the simple average of the actual values, the ratio is one. Smaller ratios indicate better performance, and a ratio of zero implies that the predictions equal the actual values. With the exception of foundation giving, the ratio is less than one for all sources of giving. Among the recipient subsectors, only publicsociety benefit giving has a ratio above one. RMSE (1) Standard Deviation (2) Ratio (1) (2) TOTAL GIVING BY INDIVIDUALS/HOUSEHOLDS GIVING BY CORPORATIONS GIVING BY FOUNDATIONS GIVING BY ESTATES EDUCATION GIVING HEALTH GIVING PUBLIC-SOCIETY BENEFIT GIVING

13 Table 7 SUMMARY STATISTICS FOR THE GIVING VARIABLES AND EXPLANATORY VARIABLES USED IN THE MODELS Variable Average Standard Deviation Min Max Variable Average Standard Deviation Min Max G C S E N T GHEALTH GCPROF GHEALTHCARE SERVICES G C S A V E GSERVICES HIGHERED D C T A X GGOODSJEWELRY ANDWATCHES GDPINC GGDOODSMOTOR VEHICLES G E M P GGOODSNEW MOTORVEHICLES GGDP GNURSERYTOHS GNWORTH GSERVICESFINE XPNPO TAXDUM GNPORECEIPT SALES GNITEM NETNPOSERVICES DPTAX GPHARMA DR1T GRECREATION GMBASE GRECSERVICES GCON GSOCIALAND RELIGSERVICES PCTHEALTHTOGDP GGOODSEDUC BOOKS GCLOTHING GPINC GCOMMUNITY SCHOOLS GPSAVE GEDUCATION DPSRATE GEDUCSERVICES DRECM GFOREIGNTRAVEL GSP GFURNISHINGS THE PHILANTHROPY OUTLOOK 2019 &

14 Figure 1 ACTUAL VS. PREDICTED GROWTH RATES FOR TOTAL GIVING, Figure 1 provides a comparison of forecasted growth rates against actual growth rates for total giving. The average and standard deviation of the forecasts match the actual average and standard deviation quite closely. The forecast average is 0.64%, and the actual average is 0.74%. The standard deviation of the forecast is 5.04%, versus 5.25% for the actual. However, as Figure 1 shows, the forecast model missed a large spike in the giving growth rate in The predicted spike came in the following year when the actual growth rate had fallen below zero. The model did capture the downturn in the giving growth rate in 2008 at the onset of the Great Recession. 12

15 1 Also referred to as explanatory variables. 2 There are several terms used for the various personal consumer expenditures in The Philanthropy Outlook 2019 & 2020 models. Please refer to Table 1b to view these terms. 3 Chapter 5: Personal Consumption Expenditures, Concepts and Methods of the U.S. National Income and Product Accounts, Bureau of Economic Analysis, U.S. Department of Commerce, November 2017, 4 Predictions are based on annual data available in October Predictions are based on annual data available in October Predictions are based on annual data available in October Predictions are based on annual data available in October Predictions are based on annual data available in October Predictions are based on annual data available in October Please note that while the giving projections based on 10 th percentile GDP estimates form the lower bound of the baseline giving projections, the opposite is true of bequest giving, for which they form the upper bound. The counterintuitive direction of the giving projections based on the 10 th and 90 th percentile estimates results from the difficulty associated with predicting bequest giving from year to year and the negative feedback loop created by our bequest model. Historically, growth in bequest giving has been driven by the random chance of a very large bequest being made in a given year and a subsequent drop in bequest giving the following year. 11 Please note that while the giving projections based on 90 th percentile GDP estimates form the upper bound of the baseline giving projections, the opposite is true of bequest giving, for which they form the lower bound. The counterintuitive direction of the giving projections based on the 10 th and 90 th percentile estimates results from the difficulty associated with predicting bequest giving from year to year and the negative feedback loop created by our bequest model. Historically, growth in bequest giving has been driven by the random chance of a very large bequest being made in a given year and a subsequent drop in bequest giving the following year. THE PHILANTHROPY OUTLOOK 2019 & THE PHILANTHROPY OUTLOOK 2019 &

16