### Table 1: Information on time-varying test data sets.

### Table 4 Time-Varying Market Price of Currency Risk

"... In PAGE 25: ... To the extent that #20 and #20 #03 are time-varying, or that the correlation #1A zz #03 is time-varying, the sign of the currency risk premium may also be time-varying. Table4 allows the market price of currency risk to depend on the level of the exchange rate #28#20 t = #20 0 + #20 1 e t , model B#29, the interest rate di#0Berential #28#20 t = #20 0 + #20 2 #28r t , r t #03 #29, model C#29, or the volatility of the exchange rate #28#20 t = #20 0 + #20 3 v t , model D#29. Each line in the table presents only estimates of #20 0 , #20 1 , and #20 2 , along with the resulting log likelihood of the model.... In PAGE 25: ... However, when we let the market price of currency risk depend on both the level and the volatility of the exchange rate #28model E#29, only the dependence on the volatility remains signi#0Ccant. Plot A of Figure 5 shows a decomposition of the exchange rate drift with a time-varying market price of currency risk #28model D in Table4 #29. The solid line is the interest rate di#0Berential, the dashed line is the currency risk premium, and the dotted line is the interest rate risk premium.... In PAGE 26: ... Studies by Baillie and Bollerslev #281989,1990#29, Bekaert and Hodrick #281993#29, and Domowitz and Hakkio #281985#29, #0Cnd only weak support for the inclusion of the conditional exchange rate volatility in the exchange rate drift. The evidence presented in Table4 and in Figure 5 is much stronger for two reasons. We impose an economic model, which implies a speci#0Cc functional form for the drift, and we observe the instantaneous volatility of the exchange rate, rather than infer it with error from observed changes in the exchange rate.... In PAGE 28: ... 4.3 Implications for Currency Markets With time-varying market price of currency risk #28model D in Table4 #29 and time-varying correlation between innovations to the log exchange rate and innovations to its volatility #28model B in Table 6#29, our estimated model is: dr t =0:240 , 0:034 , r t #01 dt +0:047 p r t dW t ; dr t #03 =1:069 , 0:070 , r t #03 #01 dt +0:093 p r t #03 dW t #03 ; #2842#29 de t = h , r t , r t #03 #01 + #10 , 4:063 , 29:817v t #01 + , , 0:230 #01, , 0:194 #01 p r t #11 v t , 1 2 v t 2 i dt + v t dX t ; dv t =4:073 , 0:102 , v t #01 dt +0:305 p v t dY t ; where Corr 2 6 6 6 6 4 dW t dW t #03 dX t dY t 3 7 7 7 7 5 = 2 6 6 6 6 4 1:000 ,0:205 1:000 ,0:230 0:056 1:000 0:059 ,0:006 #1A xy 1:000 3 7 7 7 7 5 #2843#29 and #1A xy =2 exp #08 1:573 , 3:217e t #09 1 + exp #08 1:573 , 3:217e t #09 , 1: #2844#29 This model has some interesting implications for the currency spot and options markets. 4.... ..."

### Table 6 Time-Varying Correlation between dX and dY

"... In PAGE 27: ...he theoretical literature on option pricing under stochastic volatility #5Be.g. Hull and White #281987#29 or Stein and Stein #281991#29#5D. Table6 allows this correlation to depend on the log exchange rate #28model B#29, the interest rate di#0Berential #28model C#29, or the volatility of the exchange rate #28model D#29. Each line in the table presents only estimates of #1A 0 , #1A 1 , and #1A 2 of the speci#0Ccation #2838#29, along with the resulting log likelihood of the model.... In PAGE 28: ... 4.3 Implications for Currency Markets With time-varying market price of currency risk #28model D in Table 4#29 and time-varying correlation between innovations to the log exchange rate and innovations to its volatility #28model B in Table6 #29, our estimated model is: dr t =0:240 , 0:034 , r t #01 dt +0:047 p r t dW t ; dr t #03 =1:069 , 0:070 , r t #03 #01 dt +0:093 p r t #03 dW t #03 ; #2842#29 de t = h , r t , r t #03 #01 + #10 , 4:063 , 29:817v t #01 + , , 0:230 #01, , 0:194 #01 p r t #11 v t , 1 2 v t 2 i dt + v t dX t ; dv t =4:073 , 0:102 , v t #01 dt +0:305 p v t dY t ; where Corr 2 6 6 6 6 4 dW t dW t #03 dX t dY t 3 7 7 7 7 5 = 2 6 6 6 6 4 1:000 ,0:205 1:000 ,0:230 0:056 1:000 0:059 ,0:006 #1A xy 1:000 3 7 7 7 7 5 #2843#29 and #1A xy =2 exp #08 1:573 , 3:217e t #09 1 + exp #08 1:573 , 3:217e t #09 , 1: #2844#29 This model has some interesting implications for the currency spot and options markets. 4.... ..."

### Table 6: Comparison of augmenting method dispatch times for varying degrees of modularity

"... In PAGE 7: ...able5:Comparisonofdispatchtimesformultiplyoperation.............................. 88 Table6 : Comparison of augmenting method dispatch times for varying degrees of modularity .... In PAGE 95: ...) To measure the dispatch speed we instantiated one real, one integer, and one rational and invoked the multiply operation 1,000,000 times on each possible combination (for a total of 9,000,000 invoca- tions.) This test was repeated for both implementations; the results appear in Table6 . The table shows that the multiple dispatch and typecases approach yield the same performance.... In PAGE 96: ... This is an apples-to-oranges comparison, but it provides valuable information, particularly for choosing imple- mentation strategies when one knows that the code will not have to be extended.37 Table6 compares the performance of the pretty-print operation for trees implemented in regular Java code, using the visitor pattern, using extensible visitor, and using an external generic function in MultiJava. The intent of these tests is to compare the cost of modularity for various partial solutions to the augmenting method problem.... ..."

### Table 3: Tests for time varying parameters in periodic model speci cations

"... In PAGE 10: ... The case of German consumption data illustrates how di cult periodic time series are to specify in applied work if forecasting performance is seen as a criterion for model evaluation. Tests for periodic parameter variation for models in rst di erences gave convincing evidence in favour of a time dependent data structure (see Table3 ). In terms of ex{post forecasting non{periodic models in rst and seasonal di erences are clearly outperformed by periodic... ..."

### Table 2: Percentage of execution time for varying percentages of most relevant predicates

2003

"... In PAGE 31: ... Whether without evidence construction (Figure 4(a) or with evidence construction (Figure 4(b)), the graphs indi- cate that the timings of the programs with mode are consistently better than those without mode declaration. Additionally, we compare the running space performance between the programs with and with- out mode declaration in Table2 . For benchmarks without evidence construction, our experiments... In PAGE 32: ...39 305.19 Table2 : Running space comparison (Megabytes) indicate that with mode declaration, space requirement is 1.4 to 15.... In PAGE 54: ...564 1.577 Table2 : Normalized execution times. conj disj sh 65 gr 65 sh 130 gr 130 sh 5gr5sh10 gr 10 comp 11.... In PAGE 54: ...775 Table 3: Normalized compilation times. The results in Table2 , 3, and 4 are normalized with respect to the control flow compilation case, namely with respect to conj sh 65 for the conj benchmarks and with respect to disj sh 5forthe disj benchmarks. Table 2 shows the normalized execution time of a query when it is executed using compile amp; run (comp), meta-call (call), embedded meta-call (emc), and control flow compiled code (cfcomp).... In PAGE 54: ... The results in Table 2, 3, and 4 are normalized with respect to the control flow compilation case, namely with respect to conj sh 65 for the conj benchmarks and with respect to disj sh 5forthe disj benchmarks. Table2 shows the normalized execution time of a query when it is executed using compile amp; run (comp), meta-call (call), embedded meta-call (emc), and control flow compiled code (cfcomp). Embedding the meta-call results in a substantial improvement over normal meta-call, which is of course due to the massive instruction compression.... In PAGE 65: ...3. Table2 shows differences of about 3% in time both ways. That hardly seems meaningful, but the meta qsort and queens are very backtracking intensive.... In PAGE 66: ...6 12402 12111 +2.3 Table2 : Time (msecs) and space (machine words) performance: one stack against two stack On the whole, the one stack model is favorable to backtracking intensive programs. Note that the space figures in Table 2 include the setup for the benchmarks5.... In PAGE 66: ...3 Table 2: Time (msecs) and space (machine words) performance: one stack against two stack On the whole, the one stack model is favorable to backtracking intensive programs. Note that the space figures in Table2 include the setup for the benchmarks5. The one stack model has more chance to win space wise when the life times of choice points and environments overlap: this seems not true in more realistic programs like comp.... In PAGE 83: ... Further than that only marginal improvements would be achieved, or the code growth could even introduce some slow-downs due to caching problems. Notice that the results from Table2 show an increasing trend as the programs become larger. Considering the last 3 programs which have more than 40 predicates, the percentage of the execution time on 20% of the predicates is on average 83.... In PAGE 93: ... The degree of complexity of the low-level code is similar to the one proposed in the BAM [25]. Table2 summarizes the instructions. The Type argument which appears in several of them is intended to reflect the type of the instruction arguments: for example, in the instruction bind, Type used to specify if the arguments contain a a more complete discussion of this issue).... In PAGE 94: ... CallerImp and CalleeImp mark how caller and callee are compiled. Control ijump(X) Jump to the address stored in X jump(Label) Jump to Label cjump(Cond, Label) Jump to Label if Cond is true switch on type(X, Var, Str, List, Cons) Jump to the label that matches the type of X switch on functor(X, Table, Else) switch on cons(X, Table, Else) Conditions not(Cond) Negate the Cond condition test(Type, X) True if X matches Type equal(X, Y) True if X and Y are equal erroneous(X) True if X hasanerroneousvalue Table2 : Control and data instructions. variable (and, if this is known, whether it lives in the heap, in the stack, etc.... In PAGE 111: ... Table2 contains the timings for the benchmarks for each bb heapwb system. The maximal size of the remembered sets (number of entries) is also included.... In PAGE 111: ...3. wam heap bb heapwb 2Mb bb heapwb 4Mb bb heapwb 8Mb bb heapwb 16Mb ttot ttot mremset ttot mremset ttot mremset ttot mremset browsegc 5319 6404 1042695 6079 782547 5620 0 5668 0 boyergc 9074 9949 1115 9837 606 9769 316 9716 176 dnamatchgc 2414 2598 217 2588 120 2578 50 2571 11 takgc 1380 1465 0 1462 0 1454 0 1434 0 serialgc 7725 9114 22891761 9132 22891060 9031 22890701 9054 22395214 Table2 : Overhead of the write barrier and remembered sets browsegc boyergc dnamatchgc takgc serialgc 1.0 1.... ..."

### TABLE 2. Model Time-Varying Inputs

2007

"... In PAGE 33: ...TABLE2 . Model Time-Varying Inputs (cont.... ..."

### Table 7-1: Image Interpolation Run Times Varying Input Size 1 PE. Input Image Size

2000

"... In PAGE 83: ...represents the amount of time that the application spends performing tasks that are not computationally useful. Table7 -1 provides some performance data from the image interpolation application running an an Annapolis Micro Systems WILDFORCE board. The design runs at 25 MHz.... ..."

Cited by 2

### Table 1: Load and dump times for varying design sizes (in minutes:seconds)

"... In PAGE 16: ... We evaluated the load time and dump time for four di erent designs that represent varying degrees of complexity. These results are presented in Table1 , which shows the designs, the size of the text les that contain their parasitic information, the number of terminals, the number of signals and the corresponding load and dump times. From these results, we observe that the times taken for the transfer of data in either direction are relatively small compared to the overall time taken for all the four stages of interconnect analysis, which is typically in the order of several hours.... ..."

### Table 3. FFT Co-Processor Computation Time at Varying Clock Rates

2004

"... In PAGE 4: ...$4.95 * (13.6 / 9.06)). Table3 shows the average throughput achieved when running the EMIF and co-processor and different clock rates with pipelining. Table 3.... In PAGE 4: ... Table3 indicates that running the EMIF at higher clock rates (133 MHz rather than 100 MHz) or at a higher bandwidth (64-bit synchronous rather than 32-bit synchronous) could potentially increase the performance of the FPGA co-processor assuming that the performance bottleneck is caused by the latency in data transfer. The use of FIFO buffers in the transmit-and-receive paths enables the FFT (or any other co-processing function) to run at higher clock speeds from the EMIF.... ..."

Cited by 1